[summary of the invention]
The many aspects of the application provide obstacle detection method, device, equipment and storage medium, can be to allBarrier is all completed to divide, and will not miss barrier；Improve the real-time of detection of obstacles.
The one side of the application provides a kind of obstacle detection method, comprising:
Automatic driving vehicle ambient enviroment is scanned to obtain three-dimensional point cloud；
The three-dimensional point cloud is subjected to down-sampling, is projected on grid, grid image is constructed；
Detect the barrier in the grid image.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, it is described will be describedThree-dimensional point cloud projects on grid
The three-dimensional point cloud is subjected to down-sampling, is projected on the two-dimensional grid parallel with ground；Or,
The three-dimensional point cloud is subjected to down-sampling, is projected on 3D grid；Or,
The three-dimensional point cloud is subjected to down-sampling, before projecting on the two-dimensional grid at visual angle.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the detection instituteThe barrier stated in grid image further comprises:
Detection of obstacles is carried out to the grid image using region growing algorithm.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, it is described will be describedThree-dimensional point cloud carries out down-sampling, projects on the two-dimensional grid parallel with ground and further includes:
The peak in mean value and the direction z in each grid record cylindricality on all the points x, y direction.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, it is described will be describedThree-dimensional point cloud carries out down-sampling, projects on 3D grid and further includes:
Each grid records the mean value of all the points in the grid.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, by the three-dimensionalPoint cloud carries out down-sampling, further includes: on the two-dimensional grid at visual angle before projecting
Each grid records in the grid laser scanning point to the depth information of laser emitter.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, it is described to use areaGrid image described in the growth algorithm of domain carries out detection of obstacles, including following sub-step:
Sub-step 1, randomly selected from grid image 1 not yet ownership pixel as seed point, if the pixel is(x0, y0), with stack representation seed set, by seed point (x0, y0) push into seed storehouse；
First seed point (x0, y0) pop in seed storehouse is gone out storehouse by sub-step 2, and centered on the point, traversalThe center neighborhood territory pixel (x, y)；
Whether judgement traversal pixel (x, y) is in seed region；If not, whether judgement traversal pixel (x, y) is fullSmoothness constraint condition between foot and seed point (x0, y0)；If traversal pixel (x, y) meets condition, by the traversal pixel(x, y) push is into storehouse；Meanwhile cluster set is added in seed point (x0, y0)；
Sub-step 3 judges whether seed storehouse is empty；If so, by cluster gather in pixel constitute oneCluster, the cluster are the corresponding pixel collection of barrier, execute sub-step 4；If not, executing sub-step 2；
Sub-step 4 judges whether all pixels were all once added into seed storehouse in 2-D gray image, if so,Region increases and terminates, and exports all cluster；If not, executing sub-step 1.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, it is described to use areaGrid image described in the growth algorithm of domain carries out detection of obstacles, further comprises: according to the three-dimensional point stored during down-samplingCorresponding relationship between cloud and the pixel of grid obtains the corresponding three-dimensional point cloud of barrier.
Another aspect of the present invention provides a kind of obstacle detector, comprising: acquiring unit, projecting unit and inspectionSurvey unit；
The acquiring unit, for being scanned to obtain three-dimensional point cloud to automatic driving vehicle ambient enviroment；
The map unit projects on grid for the three-dimensional point cloud to be carried out down-sampling, constructs grid image；
The taxon, for detecting the barrier in the grid image.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the mapping are singleMember is specifically used for:
The three-dimensional point cloud is subjected to down-sampling, is projected on the two-dimensional grid parallel with ground；Or,
The three-dimensional point cloud is subjected to down-sampling, is projected on 3D grid；Or,
The three-dimensional point cloud is subjected to down-sampling, before projecting on the two-dimensional grid at visual angle.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the grouping sheetMember is specifically used for:
Detection of obstacles is carried out to the grid image using region growing algorithm.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the mapping are singleThe three-dimensional point cloud is being carried out down-sampling by member, specific to execute when projecting on the two-dimensional grid parallel with ground:
The peak in mean value and the direction z in cylindricality on all the points x, y direction is recorded in each grid.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the mapping are singleThe three-dimensional point cloud is being carried out down-sampling by member, specific to execute when projecting on 3D grid:
Each grid records the mean value of all the points in the grid.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the mapping are singleThe three-dimensional point cloud is being carried out down-sampling by member, specific to execute when before projecting on the two-dimensional grid at visual angle:
Each grid records in the grid laser scanning point to the depth information of laser emitter.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the grouping sheetMember is specific to execute when using area growth algorithm carries out detection of obstacles to the grid image:
Sub-step 1, randomly selected from grid image 1 not yet ownership pixel as seed point, if the pixel is(x0, y0), with stack representation seed set, by seed point (x0, y0) push into seed storehouse；
First seed point (x0, y0) pop in seed storehouse is gone out storehouse by sub-step 2, and centered on the point, traversalCenter neighborhood territory pixel (x, y)；
Whether judgement traversal pixel (x, y) is in seed region；If not, whether judgement traversal pixel (x, y) is fullSmoothness constraint condition between foot and seed point (x0, y0)；If traversal pixel (x, y) meets condition, by the traversal pixel(x, y) push is into storehouse；Meanwhile cluster set is added in seed point (x0, y0)；
Sub-step 3 judges whether seed storehouse is empty；If so, by cluster gather in pixel constitute oneCluster, the cluster are the corresponding pixel collection of barrier, execute sub-step 4；If not, executing sub-step 2；
Sub-step 4 judges whether all pixels were all once added into seed storehouse in 2-D gray image, if so,Region increases and terminates, and exports all cluster,；If not, executing sub-step 1.
The aspect and any possible implementation manners as described above, it is further provided a kind of implementation, the grouping sheetMember is specific to execute when using area growth algorithm carries out detection of obstacles to the grid image:
According to the corresponding relationship between the pixel of the three-dimensional point cloud and grid that store during down-sampling, barrier pair is obtainedThe three-dimensional point cloud answered.
Another aspect of the present invention, provides a kind of computer equipment, including memory, processor and is stored in the storageOn device and the computer program that can run on the processor, the processor are realized as previously discussed when executing described programMethod.
Another aspect of the present invention provides a kind of computer readable storage medium, is stored thereon with computer program, describedMethod as described above is realized when program is executed by processor.
It can be seen that based on above-mentioned introduction using scheme of the present invention, three-dimensional point cloud projected on grid, construct netTable images detect the barrier in the grid image using region growing algorithm.All barriers are all completed to divide, noBarrier can be missed, can satisfy the demand of real-time.
[specific embodiment]
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present applicationIn attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment isSome embodiments of the present application, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the artWhole other embodiments obtained without creative efforts, shall fall in the protection scope of this application.
In addition, the terms "and/or", only a kind of incidence relation for describing affiliated partner, indicates may existThree kinds of relationships, for example, A and/or B, can indicate: individualism A exists simultaneously A and B, these three situations of individualism B.SeparatelyOutside, character "/" herein typicallys represent the relationship that forward-backward correlation object is a kind of "or".
Fig. 1 is the flow chart of obstacle detection method embodiment of the present invention, as shown in Figure 1, comprising the following steps:
In 101, automatic driving vehicle ambient enviroment is scanned to obtain three-dimensional point cloud.
It in the present embodiment, can be using the laser point cloud of the barrier object on mobile lidar acquisition road.
Specifically, for vehicle obstacle recognition method operation thereon electronic equipment (such as vehicle driving electricityBrain or car-mounted terminal) laser radar can be controlled by wired connection mode or radio connection.Specifically, in vehicleIn (such as pilotless automobile) driving process, car running computer or car-mounted terminal can control laser radar to preset frequency acquisitionThe laser point cloud for acquiring the barrier object on the road of vehicle driving, to get the three-dimensional of the barrier object on roadPoint cloud data.Wherein, the point spacing on identical barrier is small, and the point spacing between different barriers is big, as shown in Figure 2.
It should be pointed out that above-mentioned radio connection can include but is not limited to 3G/4G connection, WiFi connection, bluetoothConnection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection and other currently known or exploitations in the futureRadio connection.
The specification of laser radar can use 16 lines, 32 lines or 64 lines, wherein the higher unit for indicating laser radar of line numberEnergy density is bigger, and precision is higher.In the present embodiment, using 64 line laser radars, the laser radar is by 64 laser photosensitive twoPole pipe composition detects ambient enviroment with 600r/min revolving speed, in vertical direction has 26 °~5 ° of measurement range, 100m away fromFrom the interior resolution ratio that can reach 5cm.
In some optional implementations of the present embodiment, barrier object includes: Vehicle Object, pedestrian's object.VehicleObject can include but is not limited to: bicycle, car, truck, minibus, bus.
In 102, the three dimensional point cloud is subjected to down-sampling, is projected on the two-dimensional grid parallel with ground, structureBuild two-dimensional grid image；Including following sub-step:
In 1021, cartesian coordinate is constructed；
Using laser radar as coordinate origin, using the vertical axis direction of laser radar as Z axis, using right ahead as X-axis,Y-axis is to be determined by Z axis and X-axis according to right-hand screw rule, is vehicle left.
In 1022, three-dimensional point cloud is resolved to cartesian coordinate system；
Distance and space angle according to collected each laser scanning point away from laser emitter, calculating there emerged a laser scanningCoordinate position of the point in cartesian coordinate system.For laser radar point cloud, original point cloud format is UDP network dataPacket needs to resolve its original point cloud into the cartesian coordinate system established.
During unmanned vehicle traveling, laser radar is simultaneously with certain angular speed uniform rotation, in this processIt constantly issues laser and collects the information of reflection point, to obtain comprehensive environmental information.Laser radar is collecting reflectionAlso time and the level angle of point generation can be recorded during point distance simultaneously, and each laser emitter has volumeNumber and fixed vertical angle, the coordinate of all reflection points can be calculated according to these data.Laser radar often rotates a circleThe set for all reflection point coordinates being collected into is formed a cloud.
Laser radar can measure the distance distance with object by laser reflection, because the vertical angle of laser isFixed, it is denoted as a, it is sin (a) * distance that we, which can directly find out z-axis coordinate, here.By cos (a) * distance IAvailable distance in x, the projection of y plane is denoted as xy_dist.Laser radar record reflection point apart from whileAlso the level angle b that will record lower present laser radar rotation, according to simple set conversion, the x-axis of the available point is satMark and y-axis coordinate are respectively cos (b) * xy_dist and sin (b) * xy_dist.
Preferably, the three dimensional point cloud under cartesian coordinate system is pre-processed, reserved-range in -0m < X < 50m, -Three dimensional point cloud within the scope of 25m < Y < 25m, -3m < Z < 3m.
In 1023, down-sampling is carried out to three-dimensional point cloud, is projected to along z-axis on the two-dimensional grid parallel with ground, is generatedTwo-dimensional grid image；
X, y plane in cartesian coordinate system is divided into N number of grid, the resolution ratio of grid can according to the actual situation intoRow setting；In general, one grid corresponds to the square that actual size is 0.3 meter of side length in rasterizing.
Wherein, in the mean value (x_mean, y_mean) and the direction z in each grid record cylindricality on all the points x, y directionPeak (z_max).Peak on this two-dimensional grid image recording z direction, referred to as height map.In this way, then existingAll objects being above the ground level are marked in grid, improve the robustness of detection.
Preferably, it during carrying out down-sampling to three-dimensional point cloud, stores between three-dimensional point cloud and the pixel of gridCorresponding relationship.
Preferably, the value range for adjusting all grids simulates its linear stretch to 0-255 every with the value after stretchingThe gray scale of a grid, and two dimensional gray grid image is generated with this.
In 103, the barrier in two-dimensional grid image is detected.
It is sparse discrete, a usual object (such as vehicle, pedestrian etc.) since 64 line laser radars obtain dataMany discrete fritters can be divided into, directly can not further be classified to barrier.Therefore, it is necessary to obtainedGrid image carries out correlation cluster, so that the discrete point in a barrier can flock together, to detect barrier,Obtain the corresponding image of barrier to be identified.
In the present embodiment, specifically which kind of detection model to detect that barrier can basis from the two-dimensional grid image usingDepending on actual needs, for example, being split using region growing algorithm to two-dimensional grid image.
Specifically, including following sub-step:
Step 1031, randomly selected from two-dimensional grid image 1 not yet ownership pixel as seed point, if the pictureElement is (x0, y0), with stack representation seed set, by seed point (x0, y0) push into seed storehouse；
First seed point (x0, y0) pop in seed storehouse is gone out storehouse by step 1032, and centered on the point, traversal8 neighborhood territory pixel (x, y) of center；
Whether judgement traversal pixel (x, y) is in seed region；If not, whether judgement traversal pixel (x, y) is fullSmoothness constraint condition between foot and seed point (x0, y0)；If traversal pixel (x, y) meets condition, by the traversal pixel(x, y) push is into storehouse；Meanwhile cluster set is added in seed point (x0, y0)；
Step 1033 judges whether seed storehouse is empty；If so, by cluster gather in pixel constitute oneCluster, the cluster are the corresponding pixel collection of barrier, execute step 1034；If not, executing step 1032；
Step 1034 judges whether all pixels were all once added into seed storehouse in two-dimensional grid image, if so,Region increases and terminates, and exports all cluster；If not, executing step 1031.
All pixels belong to a determining cluster, and are not connected to mutually between these cluster.
Preferably, according to the corresponding relationship between three-dimensional point cloud and the pixel of grid, barrier corresponding three can be obtainedDimension point cloud, thus, it is possible to avoid the missing inspection of obstacle object point cloud.
Further, it can be based on the corresponding three-dimensional point cloud of the barrier to be identified, identify the class of the barrierType.
Preferably, deep learning algorithm can be used, the type of barrier is identified.Which kind of deep learning algorithm specifically usedIt can be determined according to actual needs, for example, can be used using wider convolutional neural networks (CNN, Convolution NeuralNetwork) algorithm.
Convolutional neural networks are a kind of multilayer neural networks, are good at the correlation machine study that processing image is especially big imageProblem.Convolutional neural networks finally make it successfully by the continuous dimensionality reduction of the huge problem of image recognition of data by serial of methodsIt can be trained to.One typical convolutional neural networks can be made of, wherein convolutional layer convolutional layer, pond layer, full articulamentumCooperate with pond layer, form multiple convolution groups, successively extract feature, completes classification eventually by several full articulamentums.It is comprehensiveFor, convolutional neural networks are distinguished by convolution come simulation feature, and the weight for passing through convolution is shared and pond, to reduce netThe order of magnitude of network parameter completes the tasks such as classification finally by traditional neural network.
Be labeled by the type to barrier, be sent to server, so as on the server be based on LMDB orThe two-dimensional grid image in the databases such as LEVELDB and the markup information generation training sample to barrier object.It can be withUsing the training sample to using the two-dimensional grid image as input, the convolutional Neural net that the classification of barrier is identifiedNetwork is trained, and then obtains machine learning model.
Two-dimensional grid image to be identified is identified by the machine learning model after training, can determine obstacleThe type of object, such as people, bicycle, motor vehicle.
Fig. 3 be obstacle detection method of the present invention another embodiment flow chart, as shown in figure 3, include withLower step:
In 301, automatic driving vehicle ambient enviroment is scanned to obtain three-dimensional point cloud.
In 302, the three-dimensional point cloud is subjected to down-sampling, projected on 3D grid, construct 3D grid image；Including withLower sub-step:
In 3021, cartesian coordinate is constructed；
Using laser radar as coordinate origin, using the vertical axis direction of laser radar as Z axis, using right ahead as X-axis,Y-axis is to be determined by Z axis and X-axis according to right-hand screw rule, is vehicle left.
In 3022, three-dimensional point cloud is resolved to cartesian coordinate system；
Distance and space angle according to collected each laser scanning point away from laser emitter, calculating there emerged a laser scanningCoordinate position of the point in cartesian coordinate system.For laser radar point cloud, original point cloud format is UDP network dataPacket needs to resolve its original point cloud into the cartesian coordinate system established.
Preferably, the three dimensional point cloud under cartesian coordinate system is pre-processed, reserved-range in -0m < X < 50m, -Three dimensional point cloud within the scope of 25m < Y < 25m, -3m < Z < 3m.
In 3023, down-sampling is carried out to three-dimensional point cloud, is projected on 3D grid, each grid, which records in the grid, to be ownedThe mean value of point；
X, y, z space in cartesian coordinate system is divided into N number of grid, the resolution ratio of grid can be according to the actual situationIt is set；In general, one grid corresponds to the square that actual size is 0.3 meter of side length in rasterizing.
Wherein, each grid records the mean value (x_mean, y_mean, z_mean) of all the points in the grid.
Preferably, it during carrying out down-sampling to three-dimensional point cloud, stores between three-dimensional point cloud and the pixel of gridCorresponding relationship.
Preferably, the value range for adjusting all grids, by its x, y, z value linear stretch to 0-255, with the value after stretchingThe gray scale of each grid is simulated, and 3D gray scale grid image is generated with this.
Preferably, it due to the noise of laser radar itself or hanging small obstacle, is deposited in the three dimensional point cloud that can makeIn a small amount of abnormal point, therefore progress single-point filters out before grid image output and vacantly point filters out, effectively to inhibit sensorNoise and ambient noise interference.
In 303, the barrier in 3D grid image, including following sub-step are detected:
In the present embodiment, specifically which kind of detection model to detect barrier can be according to practical need using from 3D grid imageDepending on wanting, for example, being split using region growing algorithm to image.
Step 3031, randomly selected from 3D grid image 1 not yet ownership pixel as seed point, if the pixelFor (x0, y0), with stack representation seed set, by seed point (x0, y0) push into seed storehouse；
First seed point (x0, y0) pop in seed storehouse is gone out storehouse by step 3032, and centered on the point, traversal26 neighborhood territory pixel (x, y) of center；
Whether judgement traversal pixel (x, y) is in seed region；If not, whether judgement traversal pixel (x, y) is fullSmoothness constraint condition between foot and seed point (x0, y0)；If traversal pixel (x, y) meets condition, by the traversal pixel(x, y) push is into storehouse；Meanwhile cluster set is added in seed point (x0, y0)；Wherein, the smoothness constraint condition isIt traverses the distance between pixel (x, y) and seed point (x0, y0) and is less than " combined distance "；
Step 3033 judges whether seed storehouse is empty；If so, by cluster gather in pixel constitute oneCluster, the cluster are the corresponding pixel collection of barrier, execute step 3034；If not, executing step 3032；
Step 3034 judges whether all pixels were all once added into seed storehouse in 3D grid image, if so, areaDomain increases and terminates, and exports all cluster；If not, executing step 3031.
All pixels belong to a determining cluster, and are not connected to mutually between these cluster.
Preferably, according to the corresponding relationship between three-dimensional point cloud and the pixel of grid, barrier corresponding three can be obtainedDimension point cloud, thus, it is possible to avoid the missing inspection of obstacle object point cloud.
Preferably, it can be based on the corresponding three-dimensional point cloud of the barrier to be identified, identify the type of the barrier.
Fig. 4 be obstacle detection method of the present invention another embodiment flow chart, as shown in figure 4, include withLower step:
In 401, automatic driving vehicle ambient enviroment is scanned to obtain three-dimensional point cloud.
During unmanned vehicle traveling, laser radar is simultaneously with certain angular speed uniform rotation, in this processIt constantly issues laser and collects the information of reflection point, to obtain comprehensive environmental information.As laser radarScan characteristic, distance (depth) difference of adjacent laser scanning point to transmitter is smaller on same object, phase on different objectsDistance (depth) difference of adjacent laser scanning point to transmitter is larger.
In 402, by the three-dimensional point cloud carry out down-sampling, project before on the two-dimensional grid at visual angle, building before toVisual angle two-dimensional grid image；Including following sub-step:
In 4021, cartesian coordinate is constructed；
Using laser radar as coordinate origin, using the vertical axis direction of laser radar as Z axis, using right ahead as X-axis,Y-axis is to be determined by Z axis and X-axis according to right-hand screw rule, is vehicle left.
In 4022, three-dimensional point cloud is resolved to cartesian coordinate system；
Distance and space angle according to collected each laser scanning point away from laser emitter, calculating there emerged a laser scanningCoordinate position of the point in cartesian coordinate system.For laser radar point cloud, original point cloud format is UDP network dataPacket needs to resolve its original point cloud into the cartesian coordinate system established.
Preferably, the three dimensional point cloud under cartesian coordinate system is pre-processed, reserved-range in -0m < X < 50m, -Three dimensional point cloud within the scope of 25m < Y < 25m, -3m < Z < 3m.
In 4023, down-sampling is carried out to three-dimensional point cloud, before throwing by scanning sequency on the two-dimensional grid at visual angle, eachGrid records in the grid laser scanning point to distance (depth) information of laser emitter.
Preferably, it during carrying out down-sampling to three-dimensional point cloud, stores between three-dimensional point cloud and the pixel of gridCorresponding relationship.
Preferably, the value range for adjusting all grids simulates its linear stretch to 0-255 every with the value after stretchingThe gray scale of a grid, and with this generate before to visual angle two dimensional gray grid image.
The barrier in 403, before from detection to visual angle two-dimensional grid image；Including following sub-step:
In the present embodiment, specifically which kind of detection model to detect that barrier can from the two-dimensional grid image of forward direction visual angle usingIt is decided according to the actual requirements, for example, being split to preceding to visual angle two-dimensional grid image using region growing algorithm.
Specifically, including following sub-step:
Step 4031,1 is randomly selected from the two-dimensional grid image of forward direction visual angle, and there are no the pixels of ownership as seedPoint, if the pixel is (x0, y0), with stack representation seed set, by seed point (x0, y0) push into seed storehouse；
First seed point (x0, y0) pop in seed storehouse is gone out storehouse by step 4032, and centered on the point, traversal8 neighborhood territory pixel (x, y) of center；
Whether judgement traversal pixel (x, y) is in seed region；If not, whether judgement traversal pixel (x, y) is fullSmoothness constraint condition between foot and seed point (x0, y0)；If traversal pixel (x, y) meets condition, by the traversal pixel(x, y) push is into storehouse；Meanwhile cluster set is added in seed point (x0, y0)；Wherein, the smoothness constraint condition isThe depth difference traversed between pixel (x, y) and seed point (x0, y0) is less than threshold value；
Step 4033 judges whether seed storehouse is empty；If so, by cluster gather in pixel constitute oneCluster, the cluster are the corresponding pixel collection of barrier, execute step 4034；If not, executing step 4032；
Into visual angle two-dimensional grid image, whether all pixels were all once added into seed heap before step 4034, judgementStack exports all cluster if so, region, which increases, to terminate；If not, executing step 4031.
All pixels belong to a determining cluster, and are not connected to mutually between these cluster.
Preferably, according to the corresponding relationship between three-dimensional point cloud and the pixel of grid, barrier corresponding three can be obtainedDimension point cloud, thus, it is possible to avoid the missing inspection of obstacle object point cloud.
Preferably, it can be based on the corresponding three-dimensional point cloud of the barrier to be identified, identify the type of the barrier.
Down-sampling has been carried out to initial three-dimensional point cloud three-dimensional point cloud；Grid image is split using region growing algorithmIt is highly efficient compared to being split on initial three-dimensional point cloud three-dimensional point cloud, because, it is only necessary to it calculates in each neighborhood of pixel pointsOther points do not need on three-dimensional point cloud three-dimensional point cloud with k-d tree (k-dimensional tree) removal search Neighbor Points；DropLow data processing amount, improves real-time.
By the way that the three-dimensional point cloud is carried out down-sampling, project on the two-dimensional grid parallel with ground, generation has recorded zThe two-dimensional grid image of peak on direction, improves the robustness of detection.
The introduction about embodiment of the method above, below by way of Installation practice, to scheme of the present invention carry out intoOne step explanation.
Fig. 5 is the composed structure schematic diagram of obstacle detector embodiment of the present invention, as shown in Figure 5, comprising: obtainTake unit 51, projecting unit 52 and detection unit 53.
The acquiring unit, for being scanned to obtain three-dimensional point cloud to automatic driving vehicle ambient enviroment；
The map unit projects on grid for the three-dimensional point cloud to be carried out down-sampling, constructs grid image；
The taxon, for detecting the barrier in the grid image.
Preferably, the acquiring unit is used to acquire swashing for the barrier object on road in advance using mobile lidarLuminous point cloud.
The specification of laser radar can use 16 lines, 32 lines or 64 lines, wherein the higher unit for indicating laser radar of line numberEnergy density is bigger, and precision is higher.In the present embodiment, using 64 line laser radars, the laser radar is by 64 laser photosensitive twoPole pipe composition detects ambient enviroment with 600r/min revolving speed, in vertical direction has 26 °~5 ° of measurement range, 100m away fromFrom the interior resolution ratio that can reach 5cm.
In some optional implementations of the present embodiment, barrier object includes: Vehicle Object, pedestrian's object.VehicleObject can include but is not limited to: bicycle, car, truck, minibus, bus.
Preferably, the map unit is used to the three-dimensional point cloud carrying out down-sampling, projects two parallel with groundIt ties up on grid, generates two-dimensional grid image；The map unit specifically executes:
Construct cartesian coordinate；
Three-dimensional point cloud is resolved to cartesian coordinate system；
Down-sampling is carried out to three-dimensional point cloud, is projected to along z-axis on the two-dimensional grid parallel with ground, two-dimensional mesh trrellis diagram is generatedPicture；Wherein, the peak in the mean value and the direction z in each grid record cylindricality on all the points x, y direction.
Preferably, it during carrying out down-sampling to three-dimensional point cloud, stores between three-dimensional point cloud and the pixel of gridCorresponding relationship.
The three-dimensional point cloud is subjected to down-sampling, is projected on 3D grid, 3D grid image is constructed；The map unit toolBody executes:
Construct cartesian coordinate；
Three-dimensional point cloud is resolved to cartesian coordinate system；
Down-sampling is carried out to three-dimensional point cloud, is projected on 3D grid, each grid records the mean value of all the points in the grid.
Preferably, it during carrying out down-sampling to three-dimensional point cloud, stores between three-dimensional point cloud and the pixel of gridCorresponding relationship.
Preferably, the map unit for the map unit before projecting to the two-dimensional grid at visual angle, buildingForward direction visual angle two-dimensional grid image；The map unit specifically executes:
Construct cartesian coordinate；
Three-dimensional point cloud is resolved to cartesian coordinate system；
Down-sampling is carried out to three-dimensional point cloud, before being thrown by scanning sequency on the two-dimensional grid at visual angle, each grid recordDepth information of the laser scanning point to laser emitter in the grid.
Preferably, it during carrying out down-sampling to three-dimensional point cloud, stores between three-dimensional point cloud and the pixel of gridCorresponding relationship.
Preferably, the taxon, for carrying out detection of obstacles to the grid image using region growing algorithm,Detect the barrier in the grid image.
The taxon is specifically held when using area growth algorithm carries out detection of obstacles to the grid imageRow:
Sub-step 1, randomly selected from grid image 1 not yet ownership pixel as seed point, if the pixel is(x0, y0), with stack representation seed set, by seed point (x0, y0) push into seed storehouse；
First seed point (x0, y0) pop in seed storehouse is gone out storehouse by sub-step 2, and centered on the point, traversalThe center neighborhood territory pixel (x, y)；
Whether judgement traversal pixel (x, y) is in seed region；If not, whether judgement traversal pixel (x, y) is fullSmoothness constraint condition between foot and seed point (x0, y0)；If traversal pixel (x, y) meets condition, by the traversal pixel(x, y) push is into storehouse；Meanwhile cluster set is added in seed point (x0, y0)；
Sub-step 3 judges whether seed storehouse is empty；If so, by cluster gather in pixel constitute oneCluster, the cluster are the corresponding pixel collection of barrier, execute sub-step 4；If not, executing sub-step 2；
Sub-step 4 judges whether all pixels were all once added into seed storehouse in 2-D gray image, if so,Region increases and terminates, and exports all cluster；If not, executing sub-step 1.
Preferably, according to the corresponding relationship between three-dimensional point cloud and the pixel of grid, barrier corresponding three can be obtainedDimension point cloud, thus, it is possible to avoid the missing inspection of obstacle object point cloud.
Preferably, it can be based on the corresponding three-dimensional point cloud of the barrier to be identified, identify the type of the barrier.
The specific workflow of Fig. 5 shown device embodiment please refers to the related description in preceding method embodiment, no longerIt repeats.
It can be seen that based on above-mentioned introduction using mode described in above-described embodiment, down-sampling, projection carried out to three-dimensional point cloudOnto grid, grid image is constructed, detects the barrier in the grid image using region growing algorithm.It can be to all barriersHinder object all to complete to divide, barrier will not be missed, additionally it is possible to meet the needs of real-time.
Fig. 6 shows the frame for being suitable for the exemplary computer system/server 012 for being used to realize embodiment of the present inventionFigure.The computer system/server 012 that Fig. 6 is shown is only an example, should not function and use to the embodiment of the present inventionRange band carrys out any restrictions.
As shown in fig. 6, computer system/server 012 is showed in the form of universal computing device.Computer system/clothesThe component of business device 012 can include but is not limited to: one or more processor or processing unit 016, system storage028, connect the bus 018 of different system components (including system storage 028 and processing unit 016).
Bus 018 indicates one of a few class bus structures or a variety of, including memory bus or Memory Controller,Peripheral bus, graphics acceleration port, processor or the local bus using any bus structures in a variety of bus structures.It liftsFor example, these architectures include but is not limited to industry standard architecture (ISA) bus, microchannel architecture (MAC)Bus, enhanced isa bus, Video Electronics Standards Association (VESA) local bus and peripheral component interconnection (PCI) bus.
Computer system/server 012 typically comprises a variety of computer system readable media.These media, which can be, appointsThe usable medium what can be accessed by computer system/server 012, including volatile and non-volatile media, movablyWith immovable medium.
System storage 028 may include the computer system readable media of form of volatile memory, such as deposit at randomAccess to memory (RAM) 030 and/or cache memory 032.Computer system/server 012 may further include otherRemovable/nonremovable, volatile/non-volatile computer system storage medium.Only as an example, storage system 034 canFor reading and writing immovable, non-volatile magnetic media (Fig. 6 do not show, commonly referred to as " hard disk drive ").Although in Fig. 6It is not shown, the disc driver for reading and writing to removable non-volatile magnetic disk (such as " floppy disk ") can be provided, and to canThe CD drive of mobile anonvolatile optical disk (such as CD-ROM, DVD-ROM or other optical mediums) read-write.In these situationsUnder, each driver can be connected by one or more data media interfaces with bus 018.Memory 028 may includeAt least one program product, the program product have one group of (for example, at least one) program module, these program modules are configuredTo execute the function of various embodiments of the present invention.
Program/utility 040 with one group of (at least one) program module 042, can store in such as memoryIn 028, such program module 042 includes --- but being not limited to --- operating system, one or more application program, otherIt may include the realization of network environment in program module and program data, each of these examples or certain combination.JourneySequence module 042 usually executes function and/or method in embodiment described in the invention.
Computer system/server 012 can also with one or more external equipments 014 (such as keyboard, sensing equipment,Display 024 etc.) communication, in the present invention, computer system/server 012 is communicated with outside radar equipment, can also be withOne or more enable a user to the equipment interacted with the computer system/server 012 communication, and/or with make the meterAny equipment (such as network interface card, the modulation that calculation machine systems/servers 012 can be communicated with one or more of the other calculating equipmentDemodulator etc.) communication.This communication can be carried out by input/output (I/O) interface 022.Also, computer system/clothesBeing engaged in device 012 can also be by network adapter 020 and one or more network (such as local area network (LAN), wide area network (WAN)And/or public network, such as internet) communication.As shown, network adapter 020 by bus 018 and computer system/Other modules of server 012 communicate.It should be understood that computer system/server 012 can be combined although being not shown in Fig. 6Using other hardware and/or software module, including but not limited to: microcode, device driver, redundant processing unit, external magneticDish driving array, RAID system, tape drive and data backup storage system etc..
The program that processing unit 016 is stored in system storage 028 by operation, thereby executing described in the inventionFunction and/or method in embodiment.
Above-mentioned computer program can be set in computer storage medium, i.e., the computer storage medium is encoded withComputer program, the program by one or more computers when being executed, so that one or more computers execute in the present inventionState method flow shown in embodiment and/or device operation.
With time, the development of technology, medium meaning is more and more extensive, and the route of transmission of computer program is no longer limited byTangible medium, can also be directly from network downloading etc..It can be using any combination of one or more computer-readable media.Computer-readable medium can be computer-readable signal media or computer readable storage medium.Computer-readable storage mediumMatter for example may be-but not limited to-system, device or the device of electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, orAny above combination of person.The more specific example (non exhaustive list) of computer readable storage medium includes: with oneOr the electrical connections of multiple conducting wires, portable computer diskette, hard disk, random access memory (RAM), read-only memory (ROM),Erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), lightMemory device, magnetic memory device or above-mentioned any appropriate combination.In this document, computer readable storage medium canWith to be any include or the tangible medium of storage program, the program can be commanded execution system, device or device use orPerson is in connection.
Computer-readable signal media may include in a base band or as carrier wave a part propagate data-signal,Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including --- butIt is not limited to --- electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also beAny computer-readable medium other than computer readable storage medium, which can send, propagate orTransmission is for by the use of instruction execution system, device or device or program in connection.
The program code for including on computer-readable medium can transmit with any suitable medium, including --- but it is unlimitedIn --- wireless, electric wire, optical cable, RF etc. or above-mentioned any appropriate combination.
The computer for executing operation of the present invention can be write with one or more programming languages or combinations thereofProgram code, described program design language include object oriented program language-such as Java, Smalltalk, C++,It further include conventional procedural programming language-such as " C " language or similar programming language.Program code can be withIt fully executes, partly execute on the user computer on the user computer, being executed as an independent software package, portionDivide and partially executes or executed on a remote computer or server completely on the remote computer on the user computer.?Be related in the situation of remote computer, remote computer can pass through the network of any kind --- including local area network (LAN) orWide area network (WAN) is connected to subscriber computer, or, it may be connected to outer computer (such as provided using Internet serviceQuotient is connected by internet).
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of the description,The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In several embodiments provided herein, it should be understood that disclosed method and apparatus can pass through itIts mode is realized.For example, the apparatus embodiments described above are merely exemplary, for example, the division of the unit, onlyOnly a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components can be tiedAnother system is closed or is desirably integrated into, or some features can be ignored or not executed.Another point, it is shown or discussedMutual coupling, direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING or logical of device or unitLetter connection can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unitThe component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multipleIn network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme's.
It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unitIt is that each unit physically exists alone, can also be integrated in one unit with two or more units.The integrated listMember both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
Finally, it should be noted that above embodiments are only to illustrate the technical solution of the application, rather than its limitations；AlthoughThe application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it still may be usedTo modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；And these are modified or replaceed, each embodiment technical solution of the application that it does not separate the essence of the corresponding technical solution spirit andRange.