US20050026689A1 - System and method for object tracking - Google Patents

Abstract

A hand-manipulated prop is picked-up via a single video camera, and the camera image is analyzed to isolate the part of the image pertaining to the object for mapping the position and orientation of the object into a three-dimensional space, wherein the three-dimensional description of the object is stored in memory and used for controlling action in a game program, such as rendering of a corresponding virtual object in a scene of a video display. Algorithms for deriving the three-dimensional descriptions for various props employ geometry processing, including area statistics, edge detection and/or color transition localization, to find the position and orientation of the prop from two-dimensional pixel data. Criteria are proposed for the selection of colors of stripes on the props which maximize separation in the two-dimensional chrominance color space, so that instead of detecting absolute colors, significant color transitions are detected. Thus, the need for calibration of the system dependent on lighting conditions which tend to affect apparent colors can be avoided.

Description

BACKGROUND OF THE INVENTION
• 1. Field of the Invention
• The present invention relates to computer vision systems, and more particularly to a system in which an object is picked-up via an individual video camera, the camera image is analyzed to isolate the part of the image pertaining to the object, and the position and orientation of the object is mapped into a three-dimensional space. A three-dimensional description of the object is stored in memory and used for controlling action in a game program, such as rendering of a corresponding virtual object in a scene of a video display.
• 2. Background of the Invention
• Tracking of moving objects using digital video cameras and processing the video images for producing various displays has been known in the art. One such application, for producing an animated video version of a sporting event, has been disclosed by Segen, U.S. Pat. No. 6,072,504, the disclosure-of which is incorporated in the present specification by reference. According to this system, the position of a tennis ball during play is tracked using a plurality of video cameras, and a set of equations relating the three-dimensional points in the court to two-dimensional points (i.e. pixels) of digital images within the field of view of the cameras are employed. Pixel positions of the ball resolved in a given digital image can be related to a specific three-dimensional position of the ball in play and, using triangulation from respective video images, a series of image frames are analyzed by a least-squares method, to fit the positions of the ball to trajectory equations describing unimpeded segments of motion of the ball.
• As described in some detail by Segen, once a three-dimensional description of position and motion of an object has been determined, various methods exist which are well known in the art for producing an animated representation thereof using a program which animates appropriate object movement in a video game environment.
• Stated otherwise, Segen is concerned with determining the three-dimensional position of an object in motion from a plurality of two-dimensional video images captured at a point in time. Once the three-dimensional position of the “real” object is known, it is then possible to use this information to control a game program in any number of different ways which are generally known to game programmers.
• However, the system of Segen relies on a plurality of video cameras for developing positional information about the object based on triangulation. Moreover, the detected object of Segen is a simple sphere which does not require information about the orientation (e.g. inclination) of the object in space. Thus, the system of Segen is not capable of reconstructing position and orientation of an object, whether moving or at rest, from a two-dimensional video image using a single video camera.
• It is common for game programs to have virtual objects formed from a combination of three-dimensional geometric shapes, wherein during running of a game program, three-dimensional descriptions (positions and orientations) of the objects relative to each other are determined by control input parameters entered using an input device such as a joystick, game controller or other input device. The three-dimensional position and orientation of the virtual objects are then projected into a two-dimensional display (with background, lighting and shading, texture, and so forth) to create a three-dimensional perspective scene or rendition by means of the rendering processor functions of the game console.
• As an example, there can be “virtual object” that forms a moving image in a game display corresponding to how one moves around the “real” object. To display the virtual object, the calculated three-dimensional information is used for fixing the position and orientation of the “virtual object” in a memory space of the game console, and then rendering of the image is performed by known projection processing to-convert the three-dimensional information into a realistic perspective display.
• However, in spite of the above knowledge and techniques, problems continue to hinder successful object tracking, and a particularly difficult problem is extracting precisely only those pixels of a video image which correspond unambiguously to an object of interest. For example, although movement of an object having one color against a solid background of another color, where the object and background colors vary distinctly from one another, can be accomplished with relative ease, tracking of objects, even if brightly colored, is not so easy in the case of multi-colored or non-static backgrounds. Changes in lighting also dramatically affect the apparent color of the object as seen by the video camera, and thus object tracking methods which rely on detecting a particular colored object are highly susceptible to error or require constant re-calibration as lighting conditions change. The typical home use environment for video game programs demands much greater flexibility and robustness than possible with conventional object tracking computer vision systems.
SUMMARY OF THE INVENTION
• It is a principal object of the present invention to provide an object tracking system suitable for video game programs which overcomes the aforementioned disadvantages of the conventional art, enabling tracking of the position and orientation of an object or prop which serves as an input device for effecting action in the game program.
• Another object of the invention is to provide alternative interfaces for games, wherein rather than using a conventional joystick, the user can stand in front of a video camera connected to the game console device, and by physically moving or manipulating an object within view of a single camera, cause a corresponding action to occur in the game.
• Yet another object of the invention is to provide methods for mapping two-dimensional information of a discriminated pixel group belonging to the manipulated object to a three-dimensional space, to provide a three-dimensional description of the position and orientation of the object in three dimensions from a single video camera image.
• Yet a further object of the invention is to provide a three-dimensional description of an object which includes a rotational component of the manipulated object.
• A still further object of the invention is to provide techniques for the selection of object colors which maximizes one's ability to discriminate, on the basis of color transitions, the pixel groups from a video image which belong unambiguously to the manipulated object and which provide the needed information for deriving a description of three-dimensional position and orientation of the object.
• The above and other objects, features and advantages of the present invention will become apparent from the following description when taken in conjunction with the accompanying drawings in which preferred embodiments of the present invention are shown by way of illustrative example.
BRIEF DESCRIPTION OF THE DRAWINGS
• FIG. 1 is a block diagram showing an example of a configuration of a main part of a video game console which is adapted to receive an input from a digital video camera.
• FIG. 2 is an illustration showing movement of a hand held prop, as an auxiliary input device, in front of a digital camera for causing corresponding action on a video display of a game program.
• FIG. 3 is a block diagram showing the functional blocks required for tracking and discrimination of the prop as it is manipulated by the user.
• FIG. 4A illustrates a prop device according to one aspect of the present invention;
• FIG. 4B illustrates a process for mapping two-dimensional pixel data of a cylinder corresponding to the prop device shown inFIG. 4A to a three-dimensional space;
• FIG. 5A illustrates a prop device according to another aspect of the present invention;
• FIG. 5B illustrates a process for mapping two-dimensional pixel data of a combined sphere and cylinder corresponding to the prop device shown inFIG. 5A to a three-dimensional space.
• FIG. 6A illustrates a prop device according to still another aspect of the present invention.
• FIG. 6B illustrates a process for mapping two-dimensional pixel data of stripes provided on a cylinder corresponding to the prop device shown inFIG. 6A to a three-dimensional space on the basis of color transitions at the stripes.
• FIG. 7 illustrates a prop device having a helical stripe thereon, and provides a description of principles of another aspect of the present invention whereby a rotational component of the prop can be determined.
• FIGS. 8A and 8B are graphs for describing a two-dimensional chrominance color space, for illustrating principles by which color transitions associated with colored stripes provided on a manipulated object are selected to maximize their detectability.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
• FIG. 1 is a block diagram of a configuration of a main part of avideo game console 60 adapted for use with a manipulated object (prop) serving as an alternative input device.
• Thegame console60 constitutes a component of anoverall entertainment system110 according to the present invention which, as shown inFIG. 1 is equipped by a multiprocessor unit MPU112 for control of theoverall system110, amain memory114 which is used for various program operations and for storage of data, avector calculation unit116 for performing floating point vector calculations necessary for geometry processing, animage processor120 for generating data based on controls from theMPU112, and for outputting video signals to a monitor80 (for example a CRT), a graphics interface (GIF)112 for carrying out mediation and the like-over a transmission bus between theMPU112 orvector calculation unit116 and theimage processor120, an input/output port124 for facilitating reception and transmission of data to and from peripheral devices, an internal OSD functional ROM (OSDROM)126 constituted by, for example, a flash memory, for performing control of a kernel or the like, and a real time clock128 -having calendar and timer functions.
• Themain memory114,vector calculation unit116,GIF112, OSDROM126,real time clock128 and input/output port124 are connected to theMPU112 over adata BUS130.
• Further connected to theBUS130 is animage processing unit138 which is a processor for expanding compressed moving images and texture images, thereby developing the image data. For example, theimage processing unit138 can serve functions for decoding and development of bit streams according to the MPEG2 standard format, macroblock decoding, performing inverse discrete cosine transformations, color space conversion, vector quantization and the like.
• A sound system is constituted by a sound processing unit SPU171 for generating musical or other sound effects on the basis of instructions from the MPU112, asound buffer173 into which waveform data may be recorded by theSPU171, and aspeaker175 for outputting the musical or other sound effects generated by theSPU171. It should be understood that thespeaker175 may be incorporated as part of thedisplay device80 or may be provided as a separate audio line-out connection attached to anexternal speaker175.
• Acommunications interface140 is also provided, connected to theBUS130, which is an interface having functions of input/output of digital data, and for input of digital contents according to the present invention. For example, through thecommunications interface140, user input data may be transmitted to, and status data received from, a server terminal on a network. An input device132 (also known as a controller) for input of data (e.g. key input data or coordinate data) with respect to theentertainment system110, anoptical disk device136 for reproduction of the contents of anoptical disk70, for example a CD-ROM or the like on which various programs and data (i.e. data concerning objects, texture data and the like), are connected to the input/output port124.
• As a further extension or alternative to the input device, the present invention includes adigital video camera190 which is connected to the input/output port124. The input/output port124 may be embodied by one or more input interfaces, including serial and USB interfaces, wherein thedigital video camera190 may advantageously make use of the USB input or any other conventional interface appropriate for use with thecamera190.
• The above-mentionedimage processor120 includes arendering engine170, amain interface172, animage memory174 and a display control device176 (e.g. a programmable CRT controller, or the like).
• Therendering engine170 executes operations for rendering of predetermined image data in the image memory, through thememory interface172, and in correspondence with rendering commands which are supplied from theMPU112.
• Afirst BUS178 is connected between thememory interface172 and therendering engine170, and a second BUS is connected between thememory interface172 and theimage memory174.First BUS178 andsecond BUS180, respectively, have a bit width of, for example 128 bits, and therendering engine170 is capable of executing high speed rendering processing with respect to the image memory.
• Therendering engine170 has the capability of rendering, in real time, image data of 320×240 pixels or 640×480 pixels, conforming to,.for example, NTSC or PAL standards, and more specifically, at a rate greater than ten to several tens of times per interval of from {fraction (1/60)} to {fraction (1/30)} of a second.
• Theimage memory174 employs a unified memory structure in which, for example, a texture rendering region and a display rendering region, can be set in a uniform area.
• Thedisplay controller176 is structured so as to write the texture data which has been retrieved from theoptical disk70 through theoptical disk device136, or texture data which has been created on themain memory114, to the texture rendering region of theimage memory174, via thememory interface172, and then to read out, via thememory interface172, image data which has been rendered in the display rendering region of theimage memory174, outputting the same to themonitor80 whereby it is displayed on a screen thereof.
• There shall now be described, with reference toFIG. 2, an overall system configuration by which a user holding a prop object manipulates the object in front of a digital video camera, for causing an action to occur in a video game.
• As shown inFIG. 2, the prop may comprise a stick-like object which is made up of ahandle303 which is typically black in color, and a brightly colored cylinder (i.e. having a saturated color)301. A user stands in front of thevideo camera190, which may comprise a USB webcam or a digital camcorder connected to an input/output port124 of agame console60 such as the “Playstation 2”® manufactured by Sony Computer Entertainment Inc. As the user moves the object in front of thecamera190, the features of the object relating to the cylinder are picked up by thecamera190, and processing (to be described later) is performed in order to isolate and discriminate a pixel group corresponding only the cylinder. A three-dimensional description of the cylinder, including its position and orientation in three-dimensional space, is calculated, and this description is correspondingly stored in themain memory114 of thegame console60. Then, using rendering techniques known in the art, the three-dimensional description of the object is used, to cause action in a game program which is displayed on the display screen of themonitor80. For example, a virtual object, shown as a torch for example, can be moved throughout the scene of the game, corresponding to the movements of the real object made by the user. As the user changes the position and orientation of the object by moving it, the three-dimensional description of the object in thememory114, and a corresponding rendering of the object in the rendering area ofimage memory174, are continuously updated so that the position and orientation of the virtual object, or torch, on themonitor80 changes as well.
• As noted above, the essential information which must be provided is a three-dimensional description of the object, which in the case ofFIG. 2 is a cylinder. However, the image which is picked up by the camera provides only two-dimensional pixel information about the object. Moreover, it is necessary to discriminate the pixels which relate only the object itself before a three-dimensional description thereof can be calculated.
• FIG. 3 is a block diagram showing the functional blocks used to track and discriminate a pixel group corresponding to the prop as it is being manipulated by the user. It shall be understood that the functions depicted by the blocks are implemented by software which is executed by theMPU112 in thegame console60. Moreover, not all of the functions indicted by the blocks inFIG. 3 are used for each embodiment. In particular, color transition localization is used only in the embodiment described in relation toFIGS. 6A and 6B, which shall be discussed below.
• Initially the pixel data input from the camera is supplied to thegame console60 through the input/output port interface124, enabling the following processes to be performed thereon. First, as each pixel of the image is sampled, for example, on a raster basis, a color segmentation processing step S201 is performed, whereby the color of each pixel is determined and the image is divided into various two-dimensional segments of different colors. Next, for certain embodiments, a color transition localization step S203 is performed, whereby regions where segments of different colors adjoin are more specifically determined, thereby defining the locations of the image in which distinct color transitions occur. Then, a step for geometry processing S205 is performed which, depending on the embodiment, comprises either an edge detection process or performing calculations for area statistics, to thereby define in algebraic or geometric terms the lines, curves and/or polygons corresponding to the edges of the object of interest. For example, in the case of the cylinder shown inFIG. 2 the pixel area will comprise a generally rectangular shape corresponding to an orthogonal frontal view of the cylinder. From the algebraic or geometric description of the rectangle, it is possible to define the center, width, length and two-dimensional orientation of the pixel group corresponding only to the object.
• The three-dimensional position and orientation of the object are calculated in step S207, according to algorithms which are to be described in association with the subsequent descriptions of preferred embodiments of the present invention.
• Lastly, the data of three-dimensional position and orientation also undergoes a processing step S209 for Kalman filtering to improve performance. Such processing is performed to estimate where the object is going to be at a point in time, and to reject spurious measurements that could not be possible, and therefore are considered to lie outside the true data set. Another reason for Kalman filtering is that thecamera190 produces images at 30 Hz, whereas the typical display runs at 60 Hz, so Kalman filtering fills the gaps in the data used for controlling action in the game program. Smoothing of discrete data via Kalman filtering is well known in the field of computer vision and hence will not be elaborated on further.
• InFIG. 4A, a prop which is used according to the first embodiment shall be described, and inFIG. 4B a description is given which explains how three-dimensional information of the position and orientation of the prop ofFIG. 4A is derived from a two-dimensional video image thereof.
• As shown inFIG. 4A, the prop is acylindrical body301 having a single solid color attached to ahandle303 which is preferably black in color. In order to fully define the position and orientation of the object in a three-dimensional space, a position of a given point p, typically the center of the object, in the X-Y plane and a depth quantity Z (i.e. the position of the point p on the Z axis) must be determined, together with angular information of the object in at least two different planes, for example, an inclination θ of the object in the X-Y plane, and an inclination φ of the object in the Y-Z plane. The actual physical length and diameter of thecylinder301, together with knowledge of the focal length of the camera, may be used for scaling, but are not essential for programming action in a game program since the virtual object shown on the display need not be of the same length and diameter, or even of the same shape, as the prop.
• Referring now toFIG. 4B, this figure shows a two-dimensional pixel image305 of the object produced by thevideo camera190. A frontal orthogonal view of thecylindrical object301 is picked up in the video image which appears as a generallyrectangular pixel group307, however, wherein the width of the pixel group can vary along the length l thereof as a result of the object being inclined in the phi φ direction or as a result of the distance overall of the prop from the camera. It will be understood that the inclination in the phi φ direction is not directly visible in thevideo image305.
• To determine the length, center point, etc. of thepixel group307 in accordance with the geometry processing step S205 discussed above, known area statistics calculations are used. Area statistics include the area, centroid, moment about the X-axis, moment about the Y-axis, principal moments, and the angle of principal moments, which typically are used for calculating moments of inertia of objects about a certain axis. For example, to determine the moments about the X and Y axes, respectively, if each pixel making up the pixel group is considered to correspond to a particle of a given uniform mass m making up a thin homogeneous sheet or lamina, then the moments about x and y axes of a system of n such particles (or pixels) located in a coordinate plane are defined as follows:$\begin{array}{cc}{M}_x=\sum _{k=1}^n{\leftarrow }^{\mid }{\mathrm{my}}_k& \left(1\right)\\ M_y=\sum _{k=1}^n{\leftarrow }^{\mid }{\mathrm{mx}}_k& \left(2\right)\end{array}$
• The center of mass of this system is located at the point (x, y) given by$\begin{array}{cc}x=\frac{M_y}{m},y=\frac{M_x}{m}& \left(3\right)\end{array}$
• Further, assuming the lamina is of a shape having a geometric center, such as the rectangle in the case ofFIG. 4B or a circle in the case ofFIG. 5B-(to be discussed later), the center of mass of such a lamina corresponds to the geometric center. More generally, if one knows the area statistics of the pixel region and, for example, that the two-dimensional shape is a rectangle, one can directly calculate its width, height and orientation. Similar calculations are possible with circular shapes to determine the center point and radius, for example. Representative calculations for cases of rectangles and circles can be found in standard college-level calculus or physics texts.
• Because theimage305 is already taken to be in the X-Y plane, the X-Y position of the center point p can be derived directly from the image. Also, the theta θ quantity is taken directly from the image simply by knowing any line l, determined in accordance with the geometry processing step S205 described above, which runs along the longitudinal axis of-thepixel group307 corresponding to thecylinder301. Typically, a longitudinal line l passing through the center point p is used for this purpose.
• Determination of the phi φ quantity requires some additional knowledge about the pixel width w in at least two different locations w₁and w₂wherein the ratio of the width quantities w₁: w₂provides a value which can be used for determining φ. More specifically, if thecylinder301 is inclined so that the top end thereof is closer to thecamera190 than the lower end of the cylinder, then, since the lower end of the cylinder is at a greater distance from thecamera190, the pixel width quantity w₂of the image will have a greater value than the pixel width quantity w₁, and vice versa. The ratio w₁:w₂is proportional to the inclination φ of thecylinder301 in the Y-Z plane, and therefore the phi quantity φ can be determined from this ratio. Typically, for better accuracy, a plurality of equidistant measurements of pixel widths between ends of thepixel group307 are taken, and averaging is performed to determine the ratio w₁:w₂.
• Determination of the depth quantity Z can be done in different ways. However, it is important to recognize that the size and number of pixels making up the two-dimensional pixel group307 are affected both by the inclination-of the object in the φ direction as well as by the actual distance of the physical object from thevideo camera190. More specifically, as the object inclines in the φ direction, the apparent length of the object as seen by the video camera tends to shorten, so that the length l of the pixel group shortens as well. However, at the same time, as the object moves farther away from the camera along the Z-axis, the apparent size of the object overall, including its length l, also becomes smaller. Therefore, it is insufficient simply to look at the length l alone as an indicator of how far away from the camera the object is. Stated otherwise, the depth quantity Z must be determined as a function of both and φ.
• However, if the phi quantity φ has already been determined and is known, a phi-weighted value of l, which we may call l_φ, can be determined, and the pixel length of l_φ in the image, which changes as the object is moved closer or farther from the camera while assuming that φ stays constant, then can be used to determine the depth quantity Z since l_φ will be proportional to Z.
• Another method for determining depth Z is to count the total number of pixels in thepixel group307 corresponding to the object. As the object gets closer to or farther from the camera, the number of pixels making up thepixel group307 increases or decreases respectively, in proportion to the depth quantity Z. However, again, the number of pixels in thepixel group307 is also affected by the inclination in the phi φ direction, so the number of pixels N must first be weighted by phi φ to result in a weighted quantity N_φ which is used for determining the depth quantity Z based on a proportional relationship between N_φ and Z.
• Yet another advantageous method for determining the depth quantity Z is to use the average width W_avgof the rectangle, which is calculated as the sum of a given number of width measurements of the rectangle divided by the number of width measurements taken. It should be clear that the average width of the pixel group is affected only by Z and not by the phi-inclination of the cylinder. It is also possible to determine phi φ from the ratio of the total length of the pixel group to the average width (i.e. l:W_avg), and moreover, wherein the sign of the phi-inclination can be determined based on whether w₁is greater or less than w₂.
• InFIG. 5A, a prop which is used according to another embodiment shall be described, and inFIG. 5B a description is given which explains how three-dimensional information of the position and orientation of the prop ofFIG. 5A is derived from a two-dimensional video image thereof.
• The prop according to the second embodiment, similar to the first embodiment shown inFIG. 4A, comprises a cylindrical stick-shaped object, however in this case, aspherical object309 of a different color is rigidly affixed to one end of thecylinder301. In addition, although not shown, a distal end of the cylinder may be provided which protrudes just slightly and is visible from an upper end of thesphere309. As shall be explained below, thesphere309 provides a simplified means for determining the depth quantity Z and the inclination of the object in the phi φ direction, which does not require measurement of relative widths of thecylinder301, and which does not require any weighting of the length quantity by phi φ in order to determine the depth quantity Z.
• As shown inFIG. 5B, apixel group311 corresponding to thesphere309 in the image appears as a two-dimensional circle. According to this embodiment, a radius R and center point P_sof the circle are determined according to area statistics calculations which have already been discussed above. In this case, further, the total number of pixels making up thepixel group311 of the circle can be counted for giving a pixel area of the circle. It will be appreciated that the circular pixel area will increase as thespherical object309 comes closer to thecamera190 and vice versa, and therefore, since the total number of pixels in thepixel group311 making up the circle is proportional to the depth quantity Z, the value for Z can be determined thereby.
• It should also be realized that, unlike the cylinder in the previous embodiment, the shape and size of the circular pixel group are not influenced as a result of the phi φ angle of inclination. More specifically, even if the object overall is tilted in the phi direction, thesphere309 and thepixel group311 will retain their general shape and, unlike the length of thecylinder301, will not become foreshortened as a result of such tilting. Therefore, an advantage is obtained in that the total number of pixels of the pixel group making up the circle in the image can always be related proportionately to the depth quantity Z and, for determining Z, phi-weighting as in the previous embodiment is not required.
• Determination of inclination of the object in the theta θ direction is done directly from the image, just as in the previous embodiment, by determining the angle theta θ between a center longitudinal line of thepixel group307 corresponding to thecylinder301 and the Y-axis.
• Determining the angle of inclination in the phi φ direction is handled somewhat differently than the previous embodiment. More specifically, such a quantity can be determined by knowledge of the depth quantity Z, determined as described above, and by the length l between the center point of thecircle311 and the center point of thepixel group307 which corresponds to thecylinder301. For any known and fixed depth quantity Z, the length1 (as viewed from the perspective of the camera) becomes shorter as the object is tilted in the phi φ direction. Therefore, if the Z quantity is known, it is possible to determine, simply from the length l, the degree of inclination in the phi φ direction, and it is not necessary to calculate a relative width quantity or ratio of widths, as in the embodiment shown byFIGS. 4A and 4B.
• FIG. 6A illustrates a prop device according to still another aspect of the present invention.
• As in the embodiment shown inFIG. 6A, the prop itself comprises a generallycylindrical body301. In addition, three stripes S₁, S₂and S₃having a different color than the cylinder itself are provided on the cylindrical body. Preferably, the stripes S₁, S₂and S₃are each equal in width and are spaced equidistant from each other, at either end of thecylinder301 and in the center thereof.
• According to this embodiment, a pixel group making up the cylinder is extracted from the image to provide a two-dimensional line along which to look for color transitions. To determine the quantities Z, θ and φ, positions are determined at which color transitions along any line l in the longitudinal direction of thecylinder301 occur.
• More specifically, as shown inFIG. 6B, a group made up of only those pixels corresponding to a line l along the longitudinal direction of thecylinder body301, as viewed by the camera, needs to be sampled in order to determine where along the line l distinct color transitions occur. In particular, for detecting such color transitions, the chrominance values Cr and Cb which are output as part of the YCrCb signals from thevideo camera190 are detected. For reasons which shall be explained below in connection with the criteria for selecting the stripe colors, it is preferable to use a combined chrominance value D made up of a Pythagorean distance of the combined chrominance signals Cr and Cb for each color of thecylinder301 and stripes S₁, S₂and S₃, respectively, thereby defining a separation in the two-dimensional chrominance color space used by thevideo camera190, according to the following formula (1):
  D={square root}{square root over ((ΔCr)²+(ΔCb)²)}  (4)
• By selecting colors which maximize the value of D (to be explained in more detail later), it is possible to select a threshold D_tat which only color transitions above a certain separation where D>D_tare considered to correspond to the color transitions of the stripes S₁, S₂and S₃. Accordingly, the pixels along the line of the cylinder are filtered, using such a threshold, in order to find the large color transitions corresponding to the stripes S₁, S₂and S₃.
• As shown inFIG. 6B, at positions along the line l where color transitions occur, for each stripe two spikes corresponding to positions where color transitions appear can be detected, and the center point between these spikes is taken to be the position of the stripes. Once the positions of the stripes are fixed, it is then a matter of course to determine the lengths l₁and l₂between the stripes, wherein the overall length of the cylinder is determined by the sum of l₁and l₂.
• It shall next be explained how knowledge of l₁and l₂provides sufficient information for determining the quantities of Z, θ and φ, necessary for describing the position and orientation of the object in three dimensions.
• First, since the line l defined by the pixels running along the length of the cylinder has already been determined, and since the camera is assumed to face normally to the X-Y plane, the angle θ is taken directly as the angle between the longitudinal line of the cylinder and the Y axis, basically in the same manner as the preceding embodiments.
• For determining the angle of inclination in the phi φ direction, the ratio of the lengths l₁:l₂is used. For example, in the case (as shown) in which the cylinder is inclined in the direction toward thecamera190, with the upper end of the cylinder being closer to the camera than the lower end, the length l₁will appear longer to the camera190 (since it is closer) than the length l₂. It will also be appreciated that, although the apparent lengths l₁and l₂will also be affected by the overall distance Z of the object from thecamera190, the ratio of these lengths l₁:l₂will not change and therefore this ratio provides a constant indication of the inclination of thecylinder301 in the phi φ direction.
• For determining the depth quantity Z, a procedure similar to that of the first embodiment is employed, wherein a phi-weighted quantity l_φ of the total length l (l=l₁+l₂) is determined for giving Z. More specifically, the influence of the inclination angle φ on the total apparent length l of the object is first determined, and then the total length, properly weighted by the influence of φ, is proportional to the distance (or depth quantity) Z of the object from thecamera190.
• Stated more simply, φ is determined from the ratio of l₁and l₂, and once phi φ is known, the total depth quantity Z can be determined from the sum of l₁and l₂.
• There shall now be described, in connection withFIG. 7, a method for determining a rotational component of the prop. This method may be applied in conjunction with any of the embodiments which have discussed above, by further equipping the prop device with a helical stripe S_Hthereon.
• Each of the tracking methods described above can be used to obtain five of the six degrees of freedom of the objects. The only one missing is the rotation of the cylinder about its axis. Information about the rotation of the cylinder would seem difficult to obtain because cylinders are symmetric in rotation about this axis. The approach taken by the present invention to obtain this rotational component is to add a helical stripe S_Hthat goes around thecylinder301 exactly once. As thecylinder301 is rotated, the height of the stripe S_Hwill correspond to the rotation angle.
• More specifically, as shown inFIG. 7, the cylinder301 (or the cylinder-part of the prop in the case ofFIGS. 5A and 5B) includes the single helical strip S_Hthereon which goes around the object only once. Information pertaining to the helical stripe is extracted, either from the entire pixel group313 which makes up the helical stripe or by using the color transitions corresponding to the helical stripe S_H, in order to determine, using the geometry processing discussed above, a helix H which best fits to the stripe S_H.
• In addition to the helix H, a center line l of the pixel group corresponding to the cylinder is determined as described previously. Also the overall length l of the pixel group is determined.
• For obtaining a degree of rotation of the cylinder, various heights h (only h₁and h₂are shown for simplicity) each of which define the distance between one end of the cylinder and the point p where the center line intersects the helix are determined.
• As shown on the right-hand side ofFIG. 7, thecamera190 only sees one side (or orthogonal projection) of thecylinder301 at a time. Accordingly,, the helix H determined from the extracted region of the camera image determines the degree of revolution of thecylinder301. More specifically, as shown, assuming no rotation (i.e. a rotational component of 0 degrees), a center line extending from one end to a point on the helix will have a first height h₁, whereas if the object is rotated by 45 degrees, the height of the center line l between the lower end to the point where it intersects the helix H will have a shorter height h₂. The condition shown by the far right-hand side ofFIG. 7, at a rotation of 90 degrees, represents to a unique case in which the center line will intersect the helix at two points. Hence, by calculating the heights of the center line l, a component of rotation of the cylinder301 (or any other object affixed to the cylinder and rotated thereby) can be determined.
• The specific quantity used for determining rotation is the ratio of the detected height between the lower end and the point on the helix to the total length l of the pixel group. This ratio gives a number from 0 to k (where k=h_max/l) which maps directly to a range of from 0 to 360 degrees. Thus, additional information with respect to the object and orientation of thecylinder301 in three-dimensional space can be provided. Such information can be used to control the rotation of a virtual object, for example, when displayed in a game program.
• Next, with respect toFIGS. 8A and 8B, a process for selection of colors for the stripes, according to the embodiments ofFIGS. 6A and 6B shall now be described. More specifically,FIG. 8A shows a diagram of a color space defined by luminance and radial coordinates of hue and saturation. Luminance is the brightness or intensity of the color, hue is the shift in the dominant wavelength of a spectral distribution, and saturation is the concentration of a spectral distribution at one wavelength.
• By contrast,FIG. 8B shows a two-dimensional chrominance color space corresponding to the Cr and Cb chrominance output signals of the video camera. It is well understood in the art that video cameras output signals for controlling the color of each pixel making up a video image. As shown by the color wheel diagram ofFIG. 8A, color can be defined using radial coordinates corresponding respectively to hue and saturation. However, as it is needlessly complex for computerized image processing to use radial coordinates, another more useful standard for defining color is the YCrCb color definition, which is the most common representation of color used in the video world. YCrCb represents each color by a single luma component (Y) and two components of chrominance Cr and Cb. Y may be loosely related to brightness or luminance whereas Cr and Cb make up a quantities loosely related to hue. These components are defined more rigorously in ITU-R BT.601-4 (Studio encoding parameters of digital television for standard 4:3 and wide-screen 16:9 aspect ratios) published by the International Telecommunication Union. Thus, the Cr and Cb chrominance signals for each pixel are defined by Cartesian coordinates which also can be used to determine a location within the color wheel corresponding to a certain hue and saturation.
• According to the present invention, the color of the stripes S₁, S₂and S₃and the color of thecylinder301 are chosen in such a way as to maximize stripe detectability for the video camera. Color-based tracking is notorious for its difficulties due to changes in lighting, which cause the apparent color to vary. As a result, if one is attempting to detect a certain color of blue corresponding to an object, for example, under certain lighting conditions it is possible for the color of blue, as perceived by the camera, to vary to such a degree that accurate detection of the object is made difficult. In the present invention, by looking for color transitions instead of absolute colors, a more robust tracking solution can be attained. For example, in the embodiment ofFIGS. 6A and 6B, if thecylinder301 is blue and the stripes S₁, S₂and S₃are orange, if lighting conditions change, then the apparent colors will also change. However, the transition between these colors, as shown inFIG. 6B, will still be very evident.
• As discussed above, video cameras capture data using the two-dimensional chrominance color space shown inFIG. 8B. By choosing colors for the object and stripes, respectively, which have a maximal separation D in this space, it is possible to significantly enhance the detectability of the color transitions.
• More specifically, as shown inFIG. 8B, highly saturated colors of blue and orange are located at substantially diametrically opposed sides of the color wheel and are separated by a large distance D in the color space. The actual distance D can be calculated as the hypotenuse of a triangle having sides defined by ΔCr (i.e. the difference in the Cr chrominance signal values for the two colors of blue and orange) and ΔCb (i.e. the difference in the Cb chrominance signal values for the same two colors), and hence the actual distance D is the square root of (ΔCr)²+(ΔCb)², as already discussed above in equation (4).
• Although blue and orange have been described as an example, it will be appreciated that-any other color pairs, for example green and magenta, which also possess a large separation in the chrominance color space may be used. In other words, the method provides a general criteria whereby colors may be selected using their chrominance signals Cr and Cb in such a manner to maximize their separation in the color space.
• More specifically, a generally applicable method for the selection of colors, as well as for calculating distance between any two colors, is performed in such a way that the distance between two colors is calculated as a distance projected onto a certain diameter-spoke of the color wheel. First, a given diameter-spoke on the color wheel is selected having a certain angle of orientation θ. By choosing the angle of orientation of the selected diameter on the color wheel, it is possible to select the color transitions one wants to detect. For example, if green is (1, 1) and magenta is (−1, −1), the diameter of the spoke should be set at an orientation θ of 45 degrees. Then the color separation distance is calculated simply by projecting the colors onto the 45 degree line. In this manner, for the case of green and magenta, the computed distance is exactly the same as the Pythagorean distance D discussed above, however with a diameter-line orientation of 45 degrees, the distance between blue and orange is zero, because they both project to the origin. This tells us that, for a selected diameter line of 45 degrees, green and magenta are the optimal colors for detection, since they possess the maximum separation in the color space for this diameter.
• Thus, for any given diameter angle of θ, which can be chosen from 0 to 180 degrees, the separation between two colors (Cr₁, Cb₁) and (Cr₂, Cb₂) may be calculated according to equation (5) as follows:
  D=[Cr₁ cos θ+Cb₁·sin θ]−[Cr₂·cos θ+Cb₂19 sin θ]  (5)
• The distance calculation shown by equation (5) can therefore also be used for setting the threshold D_tbased on a predetermined orientation defined by the angle θ. For example, if the color transitions for the object were in fact green and magenta, the general distance calculation above can be used for threshold setting, while fixing the angle θ of this equation at 45 degrees.
• Herein have been described several methods for determining the position and orientation of a real object manipulated in front of a video camera, by mapping the two-dimensional image information of the object captured by the camera to a three-dimensional space, wherein a three dimensional description including position and orientation of the object may be used to control action in a game program.
• Although one clear example of controlling a game program is to have a “virtual object” that forms a moving image in a game display corresponding to how the “real” object is moved or positioned, it will be appreciated that the three-dimensional information can be used to control game programs in any number of different ways foreseeable to persons skilled in the art. For example, a “theremin” like musical effect can be achieved wherein changes in the position and orientation of the manipulated object could be used to influence volume, tone, pitch, rhythm and so forth of sounds produced by the sound processor. Such a musical or rhythmic sound effect can by provided in combination with visual effects displayed on the screen of the game console, for enhancing the experience perceived by the game player.
• It shall be understood that other modifications will be apparent and can be easily made by persons skilled in the art without departing from the scope and spirit of the present invention. Accordingly, the following claims shall not be limited by the descriptions or illustrations set forth herein, but shall be construed to cover with reasonable breadth all features which may be envisioned as equivalents by those skilled in the art.

Claims (20)

1. An object tracking system comprising:

an input device configured to detect two-dimensional input pixel data from a prop device; and

a multiprocessor unit configured to calculate three-dimensional position and orientation data associated with the prop device from the two-dimensional input pixel data.

2. The object tracking system ofclaim 1 wherein the multiprocessor unit further comprises a memory configured to store the three-dimensional position and orientation data associated with the prop device.

3. The object tracking system ofclaim 1 wherein the multiprocessor unit further comprises an image processor configured to execute operations for rendering the three-dimensional position and output data associated with the prop device.

4. The object tracking system ofclaim 1 further comprising a monitor for displaying action in a game program caused by rendering the three-dimensional position and output data associated with the prop device.

5. The object tracking system ofclaim 1 further comprising the prop device configured to provide input data.

6. The object tracking system ofclaim 1 wherein the input device is a camera.

7. The object tracking system ofclaim 2 wherein the memory further comprises a display rendering region.

8. The object tracking system ofclaim 2 wherein the memory further comprises a texture rendering region.

9. A machine readable medium having embodied thereon a program being executable by a machine to perform a method for tracking and discriminating pixel groups, the method comprising:

receiving pixel data from an input device;

segmenting color whereby the color of each pixel is determined and the image is divided into two-dimensional segments of colors;

defining edges of an object in the image; and

calculating three-dimensional position and orientation data of the object.

10. The machine readable medium ofclaim 9 wherein the method further comprises localizing color transitions whereby distinct color transitions are defined.

11. The machine readable medium ofclaim 9 wherein the method further comprises filtering the three-dimensional position and orientation data.

12. The machine readable medium ofclaim 9 wherein defining edges of an object employs an edge detection process.

13. The machine readable medium ofclaim 9 wherein defining edges of an object employs performing calculations for area statistics.

14. The machine readable medium ofclaim 9 wherein a definition of an edge of the object is algebraic.

15. The machine readable medium ofclaim 9 wherein the definition of an edge of the object is geometric.

16. The machine readable medium ofclaim 11 wherein the filtering comprises Kalman filtering.

17. A method for tracking and discriminating pixel groups comprising:

receiving pixel data from an input device;

segmenting color whereby the color of each pixel is determined and the image is divided into two-dimensional segments of colors;

defining edges of an object in the image; and

calculating three-dimensional position and orientation data of the object.

18. The method ofclaim 17 further comprising localizing color transitions whereby distinct color transitions are defined prior to defining the edges of the object in the image.

19. The method ofclaim 17 further comprising filtering the three-dimensional position and orientation data.

20. A system for tracking and discriminating pixel groups comprising:

means for receiving pixel data from an input device;

means for segmenting color whereby the color of each pixel is determined and the image is divided into two-dimensional segments of colors;

means for defining edges of an object in the image; and

means for calculating three-dimensional position and orientation data of the object.

Images