US20070024701A1 - Stereoscopic wide field of view imaging system - Google Patents

Abstract

A stereoscopic imaging system incorporates a plurality of imaging devices or cameras to generate a high resolution, wide field of view image database from which images can be combined in real time to provide wide field of view or panoramic or omni-directional still or video images.

Description

CROSS REFERENCE TO RELATED APPLICATIONS
• This application claims the benefit under 35 U.S.C. § 119(e) of U.S. Provisional Patent Application No. 60/594,429, filed Apr. 7, 2005, and U.S. Provisional Patent Application No. 60/594,430, filed Apr. 7, 2005, the disclosures of both of which are incorporated by reference herein.
STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT
• This invention was made under NASA Contract Nos. NNJ04JC50C and NNJ05JE77C. The government has certain rights in this invention.
BACKGROUND OF THE INVENTION
• The concept of stitching multiple camera images together in order to compose a wide field of view image is known, as is the concept of capturing multiple video signals to compose a panoramic or omni-directional image, with some stereographic functionality. See, for example, U.S. Pat. Nos. 5,703,604, 6,323,858, 6,356,397, 6,392,699, and 7,015,954 and US Patent Application No. 2003/0117488.
• There are three general techniques for capturing omni-directional and/or stereographic images. In one technique, a camera is rotated using a servo-mechanism to image a spherical area of interest. This technique suffers from three significant drawbacks. First, the speed of image capture is limited by the rotational speed of the servo-mechanism and inertia of the assembly. This can place significant performance limits on the frame-rate and shutter-rate of the system as well as the speed with which users can scan the surroundings. Second, reliance on moving elements for operation inherently possesses greater maintenance requirements and suspect reliability. Third, multiple users of such a system are constrained to view the same part of the scene simultaneously, since only one direction can be viewed at a time.
• In another technique, a single camera captures a wide field-of-view image (up to a full hemisphere) using a specially shaped optical element (usually a convex lens or mirror). This technique is actually a relatively ubiquitous method of capturing panoramic images. However, while this approach may be affordable and relatively prevalent, it also suffers from a number of significant drawbacks.
• Because the entire scene is being captured by a single CCD (or similar image capture element), the total average information per pixel is significant, causing resolution loss. More importantly, however, because this technique generally involves projecting a spherical surface onto a flat, rectangular image capture element (e.g. the CCD or CMOS chip), significant distortion is unavoidable, negatively impacting resolution. Some amount of distortion can be processed out, but the information lost due to this inefficient image capture mechanism cannot be retrieved.
• Two types of obscuration also occur using this method of imaging. The first occurs in designs where the optical element is a convex mirror, which eliminates the ability to capture the cone directly above or below the imager. The second occurs in cases where this technique is used to capture stereoscopic panoramic images, because each of the cameras obscures the other, laterally. Additionally, to use this approach, scenes must be well lit to obtain premium image quality.
• In a third technique, images from multiple static cameras that cover the omni-directional space are stitched together. This technique provides key advantages over the previous techniques. Using multiple CCDs (or other image capture elements) to capture the entire omni-directional area increases the overall resolution of the image. Also, with the widespread use of digital imagers (digital cameras, camera phones, etc.) the cost of CCD & CMOS imaging components is rapidly decreasing, increasing the affordability of this approach. Additionally, use of more cameras, each with smaller field of view lenses minimizes distortions and the associated impact on resolution. Further, statically locating each camera improves the reliability and lowers the required maintenance of the design.
• The main drawback in using this approach is the requirement it places on processing bandwidth. Simultaneously capturing and displaying high resolution, high frame rate (e.g. 30 FPS) images requires very high data bandwidths, approaching and possibly exceeding 1 GByte per second. If significant real-time video processing is also required, the bandwidth demands increase.
• Prior art devices that create an image of the spherical surroundings by stitching together the images from multiple camera images have certain drawbacks. Most of these devices do not make use of more than 11 cameras and, thus, resolution suffers. In addition, to cover the same area with fewer cameras requires wide field of view lenses, causing distortion-induced resolution loss, as described above in connection with the use of fish-eye lenses for panoramic image capture. Obscuration is an issue in most of these prior designs when it comes to stereo capture. In particular, some of the camera systems are only able to grab stereo images by placing two of their omni-directional imagers adjacent to each other. Using such a set-up to capture panoramic wide field of view scenes is problematic as there is no easy way to remove the obscuration that each imager would create when viewing laterally.
SUMMARY OF THE INVENTION
• The present invention relates to a stereoscopic imaging system incorporating a plurality of imaging devices or cameras. The system generates a high resolution, wide field of view image database from which images can be combined to provide wide field of view or panoramic or omni-directional images. Wide field of view or panoramic images include any combination of images to increase the width of the scene. A panoramic view can extend a full 360°, in which the combined images form a full circle in a plane. Images can also be combined so that the wide field of view or panoramic view extends upwardly or downwardly from a plane. An omni-directional image extends a full 4π steradians.
• Stereoscopic images are formed by locating the imaging devices with an appropriate offset in each observed direction, creating enough parallax to provide the third dimension of depth perception. The resulting left and right signals are fed to the respective left and right eyes of a person via a suitable display device to give a stereoscopic effect. The design naturally extends to maintain any desired image offset to satisfy application requirements.
• The still or video images are output to an end user or image display device, such as a head mounted display or video monitor or output to some specialized processing device to perform 3-D depth calculations, automated target recognition or some other imaging processing. In one implementation, a user can be embedded into a scene to achieve a feeling of actually being on site. The user can scan around the scene or zoom in and out. Information is broadcast without feedback to control direction, so multiple users can access the data simultaneously and can independently look in different directions.
• The image processing operations can be implemented using a variety of processing techniques, from a computer's central processor, digital signal processors, field programmable devices such as FPGAs, or application specific integrated circuits (ASICs), leading to efficient and rapid processing.
DESCRIPTION OF THE DRAWINGS
• The invention will be more fully understood from the following detailed description taken in conjunction with the accompanying drawings in which:
• FIG. 1 is a plan view of a two-camera stereoscopic imaging system according to the present invention;
• FIG. 2 is a plan view of a further embodiment of a stereoscopic wide field of view imaging system incorporating two cameras per eye and broadcast to multiple users;
• FIG. 3 is an isometric view of a stereoscopic wide field of view imaging system incorporating three cameras per eye and a pair microphones;
• FIG. 4 is an isometric view of the imaging system ofFIG. 3 incorporated into a telerobotic assembly;
• FIG. 5 is an isometric view of a camera and camera daughter board;
• FIG. 6 is a schematic view of a main processing board for receiving image signals from a pair of camera sets, processing the image signals, and transmitting the image signals to an end use device processing board for eventual wireless broadcast;
• FIG. 7 is a schematic view of an end use device processing board designed to receive a broadcast signal and output to a display device;
• FIG. 8 is a flow chart of an image blending operation in the present invention;
• FIG. 9 is a plan view of a four-camera stereoscopic panoramic imaging system according to the present invention;
• FIG. 10 is an isometric view of a stereoscopic panoramic imaging system incorporating six camera pairs and broadcast to multiple users;
• FIG. 11 is a schematic illustration of the system ofFIG. 10 illustrating obscuration of one camera by another camera;
• FIG. 12 is an isometric view of a 24 camera imaging system for capturing panoramic and elevated images for broadcast to multiple users;
• FIG. 13 is an isometric view of a 72 camera imaging system for capturing images of nearly the entire spherical surrounding;
• FIG. 14 is a schematic illustration of blockage of one camera by another camera;
• FIG. 15A is an image from the forward camera ofFIG. 14;
• FIG. 15B is an image from the rearward camera ofFIG. 14 showing partial obstruction from the forward camera; and
• FIG. 16 is a blended image ofFIGS. 15A and 15B.
DETAILED DESCRIPTION OF THE INVENTION
• The present invention provides a stereoscopic imaging system having a plurality of imaging devices arranged to capture an image. The images from each imaging device are combined to provide a wide field of view (FOV) image or a panoramic or omni-directional image that can be transmitted to an end user device, such as a head mounted display (HMD), monitor, or projection device. An imaging device is typically considered to be one or more optical stages to focus a range of electro-magnetic waves from a given field of view onto an imaging mechanism, which may be a charge-coupled device (CCD), a Complementary Metal Oxide Semiconductor (CMOS) optical sensor, traditional film, a microbolometer infrared array, or some other mechanism to capture and store optical information. An imaging device may also be referred to as a camera herein. The number of imaging devices is selected based on the application and is directly related to the pixel density of the optical sensor and the field of view captured by the optical stage(s), discussed further below.
• The basis for stereographic display in the present invention is to provide multiple perspectives of objects within the field of view. Two cameras, separated by a distance that is perpendicular to the line of sight, provide the optical parallax necessary to achieve the effect of depth. If separate images, of high enough resolution, are fed to each eye of an observer, the brain (or other processing mechanism) re-assembles the image as if the observer were actually at the camera location. An extension of this concept is to replace the camera reserved for each eye with multiple cameras. By blending the images from these multiple cameras and feeding the combined image to a dedicated eye, stereo performance is extended to a wider field of view with no sacrifice in resolution.
• FIG. 1 illustrates the concept of using two cameras to provide stereographic imaging according to the present invention. Theimaging system10 is composed of afirst imaging device12, asecond imaging device14, and mounting structures for both,16 and18 respectively. The imaging devices are oriented such that their optical axes face similar directions. It will be appreciated that some variation in the direction of the optical axes is permitted due to adjustments for trimming, focusing or performance improvement purposes. The two imaging devices are separated by a baseline distance d, indicated byreference numeral20. If the system is meant to duplicate the visual response of a human collocated with the imaging system, the baseline should be similar to the interocular distance of a human's eyes, approximately six centimeters. However, any baseline distance can be used. Larger baselines provide greater depth perception at large distances. Smaller baselines do the opposite, allowing greater resolution of depth, close-up.
• In the embodiment shown, eachimaging device12,14 is connected to a dedicated capture/processingelectronics daughter board22,24 that controls camera operation, reads out pixel data, and performs some preliminary image processing if desired. The processed signals from these capture boards communicate with amain processing board26 where additional image processing is performed. The role of the main processing board in this system is described in greater detail below. In other embodiments, a dedicated camera daughter board may not be necessary or may be tightly integrated with the imaging device, depending, for example, on the application requirements or physical size constraints.
• The images from the cameras are ultimately transmitted to one or more image display devices, such as a head mounted display (HMD)30, an external monitor (not shown), or a projection device (not shown). A single broadcast/multiple receiver model is possible, whereby any number of remote stations could tap into the video feed, similar to the broadcast of television. Either hard-wired or wireless broadcast is possible. InFIG. 1, this communication is illustrated as a series ofarcs28 to suggest the wireless broadcast of the image feed to the remote HMD station(s). Wireless is suggested here for illustration purposes only. Hard-wired implementation of the single broadcast/multiple receiver model is as easy, the most common example of such a system being simple Ethernet.
• The communication between thecamera processing boards22,24 and the mainimage processing module26 can also utilize wireless communication. However, because of the inherent bottleneck that wireless communication can present, it is generally preferably to hard-wire all local processing modules and use wireless communication only for the final signal broadcast.
• Theimage display device30 used by the end-user is characterized by a mechanism that allows independent display of images to each eye for stereoscopic viewing. A head-mounted display is commonly used for stereoscopic viewing. However, alternate mechanisms for viewing these signals are possible. For example, the use of polarizing displays in concert with corresponding polarizing filters for each eye could be used. Regardless of the mechanism used, each eye sees only the image from the appropriate left or right (eye) camera. The brain is then able to reassemble the images to provide the three-dimensional effect. The following discussion refers to use of a head mounted display as the imaging device.
• AnHMD30 typically uses either a page-flipping scheme or an independent video input scheme. In the former, the video feed is formatted so that alternating left and right eye images are transmitted and displayed to each eye sequentially. In the latter, separate video signals are displayed to each eye and the broadcast signal is composed of two separate, parallel image feeds. In either case, the field of view of the broadcast image likely exceeds that of the receiving HMD. Thus, the user is able to independently scan around the scene, viewing a smaller, cropped area of interest. This can be accomplished through the use of a simple head tracking device attached to the HMD, or by using some other navigational device (e.g. joystick, keyboard, etc.). The benefit of this becomes apparent when considering the broadcast model described above. Because each user can independently tap into the image feed, they can all simultaneously enjoy independent cropped view-ports into the overall image database.
• In the system ofFIG. 1, image resolution is achieved at the expense of field of view.FIG. 2 illustrates a plan view of an embodiment of the present invention that provides a wider field of view. As shown, theimaging system10 includes fourimaging devices12,32,14, and34, where each device is preferably controlled by a dedicatedcamera processing board22,48,24, and50, although dedicated camera boards may not be necessary in all applications. The cameras are grouped in pairs wherecameras12 and32 are identified as left-eye cameras andcameras14 and34 are identified as right eye cameras. The two sets of cameras are separated by a selected baseline distance, as described above. The fields of view of two cameras for each eye-pair may overlap with each other at a given radius. For example, inFIG. 2, the field ofview36 ofcamera12 overlaps the field ofview40 ofcamera32 at a radius R, identified byreference numeral54. Outside of this radius, there is a region of image overlap44 between the two cameras. In an alternate embodiment, the fields of view of neighboring cameras, dedicated to a particular eye, may remain unique. In this case, a blind spot exists for that particular eye, but because the scene is imaged by both eyes, any point outside of a nominal radius from the imaging device is seen by at least one eye at all times. The impact on performance of such a blind spot is minimal in the same way that the presence of the nose is processed out to a certain extent by the brain when looking to the side.
• As with the embodiment ofFIG. 1, the captured signal from the four cameras are all fed to a central image processing board orboards26. In the processing board, the images from the twoboards22 and48 are blended to form a left eye image, and the images from the twoboards24 and50 are blended to form a right eye image. Alternatively, if no processing boards are present and the image signals are routed directly from the imaging devices to the main processing board, the signals fromimaging devices12 and32 are blended to form a left eye image, and the signals fromimaging devices14 and34 are blended to form a right eye image.
• A blending algorithm, discussed further below, stitches neighboring images together in consideration of the possible presence of image overlapregions44 and46. In addition, due to the slight offsets in the position and orientation of the two cameras, additional image processing (such as scaling) may be employed to optimize the image blend. The effect of camera offset decreases the farther away the object of interest is located. Additional processing, such as the use of range finders or more sophisticated image recognition schemes, can be employed to improve this distance dependent scaling if the effect is overly noticeable.
• Theprocessing board26 can be the processor in a desktop computer, the main processor in a digital signal processing (DSP) module, or a field programmable gate array (FPGA). In specialized situations, the use of application specific integrated circuits (ASICs) can also be used. It can also be a combination of these devices to handle the processing load. A combination of dedicated hardware and software performs the video processing in any case. Preferably, a majority of the video processing is implemented via FPGA to obtain significant boosts in system speed. By using the FPGA to directly control the communication bus between the imaging device and the display device, very fast system response is possible. Fast system response leads to the ability to capture and display higher resolution and/or higher frame rate video images.
• As in the system ofFIG. 1, the processed images are ultimately broadcast28 from the processing board(s)26 to one or more image display devices, such as the head mounted displays (HMD)30 and52 shown inFIG. 2. As in the previous embodiment, this broadcast could be via wireless or hard-wired mechanisms, wireless broadcast being illustrated byarcs28 inFIG. 2.
• The concept can be extended by adding and eventually blending the signals from any number of cameras per eye.FIG. 3 shows an isometric view of the cameras from the embodiment ofFIG. 2 in which a third camera pereye54 and56 has been added to provide a wider field of view to the user.FIG. 4 illustrates this embodiment integrated into, for example, a robotic system. The system architecture provides significant functional design flexibility, including the number and arrangement of cameras. While designing a system for use by humans, the ability to provide a field of view to meet or just exceed the mechanics of human vision should be adequate for many situations. Humans with two functioning eyes typically have a horizontal field of view of nearly 180° and a vertical field of view of about 140°. The visual acuity of a human with excellent vision, such as a fighter-aircraft pilot, is typically around 2 arc minutes per pixel.
• The present invention also provides significant product design flexibility to allow optimization of typical cost versus performance trade-offs. For example, regarding the performance of the embodiment shown inFIG. 3, suitable cameras may provide a distribution of 768 pixels horizontally and 1024 pixels vertically. By using a lens with a 51.2° vertical field of view and a 38.4° horizontal field of view, assuming no image overlap between camera fields of view, the overall system provides the user a horizontal field of view of approximately 115° horizontally and 51° vertically at a resolution of 3 arc minutes per pixel.
• While this design may be adequate for certain applications, other applications may require either better resolution or greater fields of view or both. This can be achieved by using some combination of higher pixel density imaging devices, more imaging devices, and/or lenses with varying fields of view. For example, to boost the resolution of the above embodiment to 2 arc minutes per pixel, one could implement higher performance image devices with at least 1152 horizontal pixels by 1536 vertical pixels. The particular combination of number of cameras, pixel density, and lens fields of view are chosen to satisfy the resolution, reliability, and budgetary requirements of the desired application.
• Another factor to consider is that typical head-mounted displays are limited in the field of view that they can display. For example, typical high-end, high cost, off-the-shelf models offer at most 60° diagonal fields of view, while more typical, moderately priced models offer only a 29° diagonal field of view. Thus, even the highest end HMDs are able to display only a small region of the image broadcast from the above embodiment. However, through the use of, for example, a head-tracker and local processing electronics, the user can scan the entire imaged region independent of the other users, by turning his or her head and “looking around,” discussed further below.
• To heighten the immersive effect in any of these designs, directional audio can be added.FIG. 3 shows aleft microphone58 and aright microphone60, which can be added to capture directional auditory information. These audio signals can be fed to theprocessing board26 and broadcast with the video feed to head mounted ear-phones to provide surround sound. Microphone position and number of microphones can also be varied to maximize performance.
• The individual cameras are preferably fixed to their mounting structures, so that the amount of image overlap can be optimized and need not be adjusted by the user. The cameras can, however, be adjustably mounted if desired for a particular application. For example, mechanical pitch, yaw, and roll adjustments can be implemented. Also, the cameras can be mounted to their mounting structures with an adjustable system that allows optimization of the overlap during post-manufacture trimming and that can be “locked in” position once the cameras' orientations are set. For example, the cameras can be mounted with Belleville washers to allow such adjustment.
• Focus of the individual cameras in the imaging system is addressed depending on the type of camera. Some cameras include an electronic focusing capability, while others include a manual focus. In the latter case, if the aperture is set such that the depth of field is relatively small, the ability to focus on objects at varying distances can be limited. In a passive approach to this situation, the aperture size is reduced to get a very large depth of field and the camera's gain is boosted to compensate for the reduced light entry. This can lead, however, to excess noise in the camera that could affect image quality. In an active approach, range information can be used in conjunction with optical adjustments (aperture, focus, etc.) to maximize imaging performance. The disadvantage of this approach is that cameras with adjustable optics are typically more bulky than cameras without. Thus, the particular camera is selected based on the desired application.
• FIG. 5 illustrates an embodiment of a compact dedicated camera daughter board, for example, aboard22 or24. In this embodiment, an image sensor such as12 in animage sensor housing64 is mounted directly on the board. A field programmable device (for example, FPGA or CPLD)66 is attached to the board for reading out pixel data and performing some preliminary image processing if desired. A connector (for example, SATA or PATA)68 is employed for data transfer to the main processor board. This camera daughter board is small and compact to facilitate close packing of multiple cameras.
• An exemplarymain processor board26 is illustrated schematically inFIG. 6 for a high-resolution, wide field of view, stereographic imaging system. In the embodiment shown, two sets of three cameras are provided, one set72 to simulate a right eye view and theother set74 to simulate a left eye view, such as described above in connection withFIG. 3. The board includes a number of devices, preferably implemented as programmable integrated circuits, discussed further below, to provide real time processing. In the embodiment illustrated, the devices include acamera input device82, a number of image processing devices, such as aframe rotation device84 and afield integration device86, adata compression device88, and anencoding device90 for transmission viainterface92. A circular frame buffer, a multiplexedrandom access memory78 in synchronization with asuitable memory80, such as SDR SDRAM, conveys the frames in a sequential manner to the various devices. The operations of the various devices are preferably implemented via FPGAs to improve processing speed while still allowing some programming flexibility. Application specific integrated circuits (ASICs) could alternatively be employed to implement processor operations.
• The main processor board processes frames sequentially, indicated schematically bypipeline76. Eight frames are illustrated as simultaneously in process, but the FPGAs can be sized and programmed to handle any suitable number of frames. In operation, signals from the camera daughter boards associated with the cameras are transmitted in any suitable manner to themain processor board26. For example, a low voltage differential signaling (LVDS) interface is used in the embodiment shown. The first frame is input to thecamera input device82, which places the frame on the RAM MUX to convey the frame to the next device, which in the embodiment shown is theframe rotation device84. To conserve space on the daughter board, it may be desirable to mount the cameras on the daughter board at a 90° orientation from horizontal. Thus, to provide the user with a properly oriented image, the frame must be rotated 90° back. If the cameras were oriented without this 90° rotation, this rotation operation would not be necessary. The frame rotation device then places the rotated frame back on the RAM MUX.
• Other operations can be incorporated here as desired. For example, per pixel corrections can be made, brightness, color, contrast, integration time, shutter speed, white balance, signal gain, saturation level, and gamma correction can be adjusted. In addition, localized mathematical operations, including edge detection, linear filtering, and background enhancement can be performed here with minimal overhead. In general, any operation that requires the manipulation of individual pixels or small neighborhoods of pixels can be implemented here.
• Next the frame is conveyed to thefield integration device86. At this step any further desired image processing occurs. For example, multiple images are combined into a single field, discussed further below. Other operations, such as cropping, panning, and zooming of the image can occur here. In general, any whole-image affine, quadratic, Euclidean, or rotational transformation can be performed in this step.
• The frame is then conveyed to the videodata compression device88 for compression prior to transmission to the end use device. Any suitable video data compression scheme may be used, such as a discrete cosine transform. The frame is then encoded for transmission to the end use device using any suitable transmission protocol, such as a Linux TCP stack for wireless transmission, at atransmission encoding device90 and transmitted viainterface92.
• In this manner, the processor of the present invention is capable of rapid real time processing of the image data. It will be appreciated that the devices of the processor can process the image data serially, in parallel, or both.
• Aprocessor board94 is provided at the end use display device as shown inFIG. 7. The operation of this board is also preferably implemented using an FPGA. A circular frame buffer, such asRAM MUX95 andSDR SDRAM96 is used to process frames serially, as described above. The signals from the main processor board are received byinterface92aand decoded at a wirelessframe decoding device97. The frame is decompressed at adecompression device98. At the HMD stereoframe output device99, the frames are prepared for transmission in the format used by the HMD, such as page flipping or via separate feeds to each eye, as known in the art. The frames are then transmitted to the end use device, such as the HMD or an external monitor or projection device viamonitor output element100 or both.
• FIG. 8 illustrates an example of blending two images into a single image, performed, for example, at fieldintegration circuitry element86. First, two images are read, atstep401. The edges of the two images are detected (step403), and overlapping pixels are identified (step405). The region of overlapping pixels is divided into smaller rectangles (step407). A rectangle of overlapping pixels is selected (step409). A correlation algorithm, using a parameter such as brightness or intensity, is run on the rectangles to find common pixel coordinates in the two images (step411), and the corresponding pixel coordinates are saved (step413).Steps409 through413 are repeated until the overlapping area has been covered (step415).
• The saved pixel coordinates are sorted by correlation and unreasonable points are discarded (step417). Amatrix transforming image1 coordinates to image2 coordinates is formed (step419), so that all images can be identified using a single coordinate system.Image2 coordinates are mapped intoimage1 coordinates (step421) and the mappedimage2 is attached to image1 (step423). The overlapping area is blended (step425). Known blending algorithms can be used, for example, that weight the pixels in the overlapping area from one edge to the opposite edge. Then, the combined image is output (step427) to the display or to the next step in the process.
• Many existing telerobotic systems or other end use devices incorporate low gain servo-systems to provide head and body rotation allowing, for example, the operator to view the surrounding scene by driving head rotations with the output from a head tracker. Because of the inertia of the system and the nature of the feedback controller, however, the servo gains and thus rotation rates are kept low to avoid the onset of instability. The difference in the rotational rate commanded by the turning of the user's head and the actual rate of the servo-system feels unnatural to the user and affects performance in carrying out operations.
• The extra wide field of view available from the present invention allows the insertion of an intermediate stage between the output of the head tracker and the input to the servo system. When the user's head turns, the image can immediately pan, up to the response rate of the head tracker (e.g., 100 Hz), and allow the user to look out into the peripheral areas without the servos activating. The servos can then take whatever time is necessary to recenter the head to the direction of observation, while the system rigidly maintains the user's observation direction. Thus, the user can instantaneously view a very wide field of view, unimpeded by the time constant of the servo-system. Panning is accomplished by cropping the image to a determined size with known dimensions, such as the size required for the end use device. The coordinates of a selected pixel, such as that in the upper left corner of the image, are identified in memory using a pointer. To pan, only the coordinates of the selected pixel need to be changed. Knowing the dimensions of the image, it can be quickly streamed out to the display.
• In another aspect, a wider field of view can provide the ability to electronically scale or zoom the image. The image can either be zoomed in displaying all captured pixels, providing the highest resolution, but narrowest instantaneous field of view (although accessible by panning), or the image can be zoomed out, by averaging a group of neighboring pixels into one pixel, to display the entire available field of view of the camera system within the field of view of the HMD. Although only having, for example, 60° “apparent” field of view within the HMD, the scene displayed could actually cover twice as much, at the cost of resolution. This mode is useful when a wider perspective on a scene is of greater importance than resolution, for instance, in a mobile system in which peripheral information is important. Such a system can be implemented with no modification to the existing servo controller.
• The wide field of view imaging system can also be used with a wide field of view head-mounted display, providing an instantaneous display. The peripheral information is placed into the peripheral view of the observer, embedding the user into the scene.
• The stereographic effect may be achieved in a panoramic or omni-directional imaging system as well. To introduce the governing principal of the present invention, first consider the capture of a purely horizontal wide field of view mosaic.FIG. 9 shows the top view of a simplified embodiment of the present invention employing four imaging devices or cameras. As shown, theimaging device110 is composed of afirst imaging device112, asecond imaging device114, athird imaging device116 and afourth imaging device118. In this design, the cameras are grouped in pairs (one pair for each angular orientation), where each pair consists of a left and right eye camera in order to provide the stereographic effect. For example,cameras112 and114 form a pair facing generally in the same direction, andcameras116 and118 form a pair facing generally in the same direction. The fields of view of neighboring cameras (e.g.,cameras112 and116) may or may not overlap. Such regions ofoverlap128 and130 are identified inFIG. 9.
• Each camera is mounted with an offset from acentral hub132 by an offsetarm134. The cameras are oriented such that their optical axes face substantially tangentially to a circle about the central hub. It will be appreciated that some variation in the tangential direction of the optical axes is permitted due to adjustment for trimming, focusing, or performance improvement purposes. The central hub can be used to house all processing electronics, may be hollowed out to minimize weight, and may incorporate mounting points to provide adequate support for the system.
• Each imaging device orcamera112,114,116,118 connects to an associated dedicated capture/processing electronics board136,138,140,142 that controls camera operation. The four processed signals from these capture boards communicate with acentral processing board144 where the images from the twoboards136 and140 are blended to form a left eye image and the images from the twoboards138 and142 are blended to form the right eye image. The blending algorithm may make use of the regions ofimage overlap128 and130 between cameras to stitch neighboring images together. In addition, due to the offset in position of the two cameras, additional image processing (such as scaling) may be employed to optimize the image blend. The effect of camera offset decreases the farther the object of interest is located. Additional processing, such as the use of range finders or more sophisticated image recognition schemes, can be employed to improve this distance dependent scaling if the effect is overly noticeable.
• Theprocessing board144 can be the processor in a desktop computer, the main processor in a digital signal processing (DSP) module, or a field programmable gate array (FPGA). In specialized situations, the use of application specific integrated circuits (ASICs) can also be used. It can also consist of a combination of these devices to handle the processing load. Dedicated hardware and software performs the video processing in any case. Implementing a majority of the video processing via FPGA significantly boosts system speed. By using the FPGA to directly control the communication bus between the imaging device and the display device, as described above in connection withFIGS. 6 and 7, very fast system response is possible. Fast system response leads to the ability to capture and display higher resolution and/or higher frame rate video images.
• The processed signals(s) are then communicated to one or more image display devices, such as the head mounted display (HMD)146. As with the wide field of view embodiment discussed above, either hard-wired or wireless broadcast communication is possible.
• The communication between thecamera processing boards136,138,140,142 and the mainimage processing module144 could also utilize wireless communication. However, because of the inherent bottleneck that wireless communication can present, it is generally preferably to hard-wire all local processing modules and use wireless communication only for the final signal broadcast.
• As noted above, the imaging system can transmit images to multiple users. Preferably, a wide field of view image is transmitted to each user, and the multiple display devices each include a local processor operative to pan, crop, and zoom the image as described above. In this manner, each user can view that portion of the image that is of interest to the user. Such a local processor can be operative as described above in connection with the devices of theprocessor26 ofFIG. 6. Alternatively, the display device can transmit a signal to the imaging system requesting a desired portion of the image to be transmitted to the display device.
• The embodiment illustrated inFIG. 9 can be extended to imaging wider fields of view, up to the full 360° panorama.FIG. 10 illustrates an embodiment of this invention that provides 360° stereoscopic panoramic imaging and display. The system is similar to that shown inFIG. 9 except that instead of two camera pairs112,114,116,118, four more camera pairs113,115,117,119121,123,125,127 are used, arranged tangentially about a central hub. The arrangement of all neighboring cameras is identical, providing a common parallax effect no matter which direction is being observed. As in the design ofFIG. 9, each camera has dedicated electronics that govern its operation and all signals are eventually brought to the central processing unit (or units)144 to blend and process images forbroadcast148 to theremote display stations146, similar to the processing steps described in connection withFIGS. 6 and 7 above. The benefit in allowing separate end-users to access independent cropped regions of the broadcast video database becomes more apparent in this example. For instance, using thesame display hardware146 and150, two users would be able to access the same feed but look in independent directions.
• For some applications, stereoscopic imaging may not be necessary. In this case, one set of cameras (such ascameras112,116,113,123,127, and119) can be arranged about the hub, each facing a direction tangential to a circle about the hub. This arrangement provides full panoramic coverage around the hub. Similarly, a portion of this set of tangentially arranged cameras can be provided. A single wide field of view image or a full panoramic image can be output to a display device, as described above. Alternatively, only a desired set of all the cameras illustrated inFIG. 10 can be used to provide the desired image data.
• To heighten the immersive effect, aleft microphone152 and aright microphone154 can be added to capture directional auditory information. These audio signals can be fed to theprocessing board144 and broadcast with the video feed to head mounted ear-phones to provide surround sound. Multiple microphones can be distributed over thehub132 if performance benefits. Likewise, microphone position can be varied to also maximize performance.
• As with the wide field of view embodiment discussed above, the present imaging system provides significant design flexibility for use in optimizing typical cost versus performance trade-offs. The angular offset and associated field of view of each camera is chosen to satisfy resolution, reliability, and budgetary requirements. For example, using more cameras with correspondingly smaller fields of view provides a higher resolution image with minimum distortion. However, this also increases image processing demands and overall cost. The balance struck between these conflicting criteria is governed by the required performance of the particular embodiment. For example, using a hexagonal arrangement similar to that shown inFIG. 10 requires the use of 12 cameras to capture the full 360° stereoscopic panorama. Assuming each camera has a field of view of 66° and 1024 horizontal pixels, this design provides a resolution of approximately 3.9 arc minutes per pixel. For reference, the visual acuity of the human eye is on the order of 2-3 arc minutes per “pixel”. Thus, such a design may appear grainy to the user.
• If a particular application must have a resolution of 2 arc minutes per pixel or better, two possible modifications of the above design that can provide this are as follows. Image capture devices (e.g. CCDs) that provide at least 1980 pixels in the horizontal plane can be used, assuming a lens field of view of 66°. Alternatively, the number of cameras can be doubled from 12 to 24, narrow field of view lenses (˜33°) can be used, and a 12-sided central hub and associated camera control electronics can be provided. In either of these high-resolution designs, the same amount of image information will be processed, so the processing demands are the same, although, depending on commercial considerations, one design may be more cost effective than the other. The design adopted depends on minimizing some combination of cost and complexity, and/or maximizing reliability.
• A complicating factor in the camera arrangement is that each camera obstructs to some extent the view of the neighboring camera. The present system overcomes this obstruction by using the redundant information in the overlapping fields of view to minimize the effect of this obscuration. In effect, the pixels in the forward camera that correspond to the obscured pixels of the rear camera could be swapped/blended into the rear camera image (scaled and image processed as needed) to overcome the effect of the obscuration, discussed in more detail below. Unfortunately, however, due to the finite extent of the forward camera assembly, there will always be a “blind” region that cannot be captured by this design. This is illustrated inFIG. 11, which shows a top view of a representative embodiment, focusing on the obscuration of one right eye camera. In this figure, the field ofview126 of therear camera118 is unshaded, the field ofview122 of theforward camera114 is lightly shaded. The overlappingregion130 and the effectiveblind region156 are both darkly shaded. As shown, the field of view ofcamera118 is obscured to some extent by theforward camera assembly158 holdingcamera114 and the rearward facing (left eye)camera123. The blind region is triangular in nature and is defined by three elements. The first is thefront camera assembly158, the second is theoutermost ray162 of the front camera's field of view, and the third is thelight ray164 leaving the rear lens and just intersecting the outermost corner of the front camera assembly. (In practice, it is preferable to orient each camera such that thislight ray164 corresponds to the innermost ray of each camera's field of view, since any image data inside this ray is largely obscured in-plane). The intersection of these last two rays defines an effective minimum radius R, indicated byreference numeral166, past which no blind spot is observed in the panorama. Inside this radius, there will always be a blind spot. The extent of this blind region can be reduced, however, by minimizing the size of the camera and mounting hardware as well as making use of wide field of view lenses (to the extent possible without adding excessive distortion to the image).
• This concept can be extended out-of-plane.FIG. 12 shows one possible embodiment of such a design. As in the previous in-plane design, sixcamera stations212,214,216,218,220,222 extend out from acentral hub224. However, instead of a single camera in each direction, two cameras, oriented at different elevations are used. As with the previous design, the output signal from each camera is fed to acentral processing unit244 where similar blending and processing steps are taken and the resulting video database is broadcast248 tomultiple HMD stations246 and250 for viewing. Again, at any time the HMD would only display a small region of the broadcast image, but through the use of, for example, a head-tracker and local processing electronics, the user is able to scan the entire spherical imaged region independent of the other users. As with the panoramic imaging system described above, microphone signals can be captured and broadcast to heighten the immersive effect.
• The above system makes use of 24 cameras to capture approximately 70% of the spherical surroundings. Alternate designs can provide even greater coverage of the surroundings. For example, thesystem310 shown inFIG. 13 uses 72 cameras to capture up to 96% of the spherical surroundings. As discussed previously, by trading the number of cameras, lens field of view, and imaging element pixel density, the design space of the present invention is wide enough to satisfy almost any desired resolution and/or cost requirement.
• As noted above, the stereo field of view can be extended to 360°, and in fact to a full 4π steradians, without obstructing any of the images. This is accomplished by feeding data available in the obstructing camera to the camera being obstructed.FIG. 14 illustrates a geometry in which one camera partially blocks the view from another camera, andFIGS. 15A and 15B illustrate two images from these two cameras. The image on the right is from the right camera that is partially obstructed by the camera on the left. The operation of blending two images discussed above in connection withFIG. 8 can be used to blend the images from the two cameras to remove the obstruction, as seen inFIG. 16. In a panoramic arrangement in which the cameras provide a full 360° image, the images from the first and last cameras must align, so the individual transformations between each image pair become connected. Apart from this constraint, the process of blending the first and last images is essentially that described above.
• The present invention is advantageous for a number of reasons. The capture and display of stereo panoramic and/or omni-directional still and video images offers a seamless mosaic of the surrounding scene. The camera groups can be oriented similarly to the make-up of the human user, which provides a more realistic stereoscopic (three-dimensional) immersive effect for human users. Similarly, the orientation of the cameras tangentially to a circle defining a panorama similarly allows the optical effect to more closely resemble the mechanics of human (and animal) vision and provides a more realistic stereoscopic (three-dimensional) immersive effect for human end-users.
• The imaging system can produce very high resolution images by capturing a scene with a suitable number of imaging devices. The imaging system can produce images with little distortion by using a suitable number of low-distortion imaging devices and optical elements. Typically, image distortion is proportional to the field of view captured by an optical element. By using more cameras with limited fields of view, the distortion of a captured image can be constrained below a maximum allowed level.
• The ability to produce stereo, wide field of view or omni-directional images without introducing moving or servo-controlled mechanisms is advantageous, because the overall reliability remains high and less maintenance is needed. Also, the present invention requires no mechanical control systems to manipulate hardware, which could lead to instability, especially over time as components age, wear, and degrade.
• In the present invention, audio surround sound can be added for presentation to the user to maximize the immersive effect.
• The present imaging system combines the capture, processing and redisplay of the optical images into a single, or multiple dedicated, processing modules, thereby providing greater speed and better software reliability. The ability to further specialize these processing modules by making use of field programmable gate arrays (FPGA), significantly boosts the processing speed of the system.
• The imaging system further allows multiple users to access the video database independently by storing and arranging the wide field of view image data in a specialized format. The ability to wirelessly transmit the captured and pre-processed video images to multiple remote HMD stations allows independent access to the image database by multiple users, simultaneously.
• The stereoscopic, wide field of view imaging and viewing system is applicable to a variety of technologies. For example, telerobotic applications, i.e., human control of robotic systems, can benefit from a very high resolution, wide field of view imaging system to enable precise motion and grasping control due to the inherent depth perception and peripheral vision provided. Applications include battlefield operations and remote surgeries. Other applications include situations or environments that may present a danger or difficulty to humans, such as searching for victims in fires, or maintenance and inspection of space platforms.
• The invention is not to be limited by what has been particularly shown and described, except as indicated by the appended claims.

Claims (38)

1. A wide field of view imaging system comprising:

a plurality of cameras arranged with overlapping fields of view; and

at least one processor comprising a plurality of devices in communication with a frame buffer operative to process image frames sequentially from a camera input device to an image transmission device, the devices including an image processing device operative to blend the image data from the cameras into a single wide field of view image and to transmit at least a portion of the single wide field of view image to an image display device.

2. The imaging system ofclaim 1, wherein the cameras are arranged in pairs, each camera in a camera pair oriented such that their optical axes face similar directions, each camera in the camera pair separated by a distance, one camera from each camera pair associated with a right eye and the other camera from each camera pair associated with a left eye;

the at least one processor operative to transmit image data associated with the right eye to a right eye display of the image display device and to transmit image data from the cameras associated with the left eye to a left eye display of the image display device to provide stereoscopic imaging.

3. The wide field of view imaging system ofclaim 1, wherein the processor comprises a daughter board associated with each camera and operative to receive data signals from the associated camera, and a main processor board operative to receive data signals from each camera daughter board.

4. The wide field of view imaging system ofclaim 1, wherein the processor comprises a plurality of devices in communication with a frame buffer operative to process image frames sequentially from a camera input device to an image transmission device.

5. The wide field of view imaging system ofclaim 4, wherein the processor is operative to process image frames serially.

6. The wide field of view imaging system ofclaim 4, wherein the processor is operative to process image frames in parallel.

7. The wide field of view imaging system ofclaim 4, wherein the devices include a frame rotation device.

8. The wide field of view imaging system ofclaim 4, wherein the devices include a device operative to perform pixel corrections.

9. The wide field of view imaging system ofclaim 4, wherein the devices include a device operative to manipulate individual pixels or neighborhoods of pixels.

10. The wide field of view imaging system ofclaim 4, wherein the devices include a device operative to adjust brightness, color, contrast, integration time, shutter speed, white balance, signal gain, saturation level, or gamma correction.

11. The wide field of view imaging system ofclaim 4, wherein the devices include a device operative to perform an edge detection operation, linear filtering, or background enhancement.

12. The wide field of view imaging system ofclaim 4, wherein the devices include a field integration device.

13. The wide field of view imaging system ofclaim 4, wherein the devices include a device operative to blend multiple images into a single image.

14. The wide field of view imaging system ofclaim 4, wherein the devices include a device operative to perform cropping, panning, or zooming of an image.

15. The wide field of view imaging system ofclaim 4, wherein the devices include a device operative to perform whole-image affine, quadratic, Euclidean, or rotational transformation operations.

16. The wide field of view imaging system ofclaim 4, wherein the devices include a data compression device.

17. The wide field of view imaging system ofclaim 4, wherein the devices include an image data encoding device.

18. The wide field of view imaging system ofclaim 4, wherein the frame buffer comprises a multiplexed random access memory in synchronous communication with a memory element.

19. The wide field of view imaging system ofclaim 4, wherein the devices comprise field programmable devices.

20. The wide field of view imaging system ofclaim 4, wherein the devices comprise field programmable gate arrays.

21. The wide field of view imaging system ofclaim 4, wherein the devices comprise application specific integrated circuits.

22. The wide field of view imaging system ofclaim 4, wherein the devices include a digital signal processor.

23. The wide field of view imaging system ofclaim 4, wherein the devices include an electronic processor.

24. The wide field of view imaging system ofclaim 1, wherein the processor is operative to transmit the image data to a head mounted display device.

25. The wide field of view imaging system ofclaim 1, wherein the processor is operative to transmit the image data to the image display device in a page flipping mode.

26. The wide field of view imaging system ofclaim 1, wherein the processor is operative to transmit the image data to the image display device separately to a right eye display and to a left eye display.

27. The wide field of view imaging system ofclaim 1, wherein the processor is operative to transmit the image data to a video monitor or a video projector.

28. The wide field of view imaging system ofclaim 1, wherein the processor is operative to transmit the image data via a wireless connection.

29. The wide field of view imaging system ofclaim 1, wherein the processor is operative to transmit the image data via a wired connection.

30. The wide field of view imaging system ofclaim 1, further comprising at least one microphone in communication with the processor to provide an audio input.

31. A method of providing a wide field of view image comprising:

arranging a plurality of cameras with overlapping fields of view;

providing at least one processor having a memory and a plurality of processing devices arranged to process frames serially from an input to an output to a display device;

inputting image data from the plurality of cameras as frames to a processor;

blending two or more frames to provide a single wide field of view image; and

transmitting the wide field of view image from the processor to the display device.

32. The method ofclaim 31, further comprising adjusting brightness, color, contrast, integration time, shutter speed, white balance, signal gain, saturation level, or gamma correction.

33. The method ofclaim 31, further comprising rotating a frame to a different orientation

34. The method ofclaim 31, further comprising cropping, panning, or zooming the image.

35. The method ofclaim 31, further comprising compressing the image data prior to transmission to the display device.

36. The method ofclaim 31, further comprising encoding the image data prior to transmission to the display device.

37. The method ofclaim 31, further comprising transmitting the wide field of view image to a plurality of display devices.

38. The method ofclaim 31, further comprising receiving the transmitted image at the display device, and providing decompression, cropping, panning, and zooming of the image at the display device.

Images