Object detection method and apparatus, electronic device, and storage medium

Abstract

The present application relates to an object detection method, device, electronic equipment and storage medium. The object detection method comprises: obtaining first point cloud data obtained by scanning a lane area corresponding to a vehicle; obtaining first road edge data obtained by performing road edge detection on the first point cloud data, performing augmentation processing on the first road edge data to obtain second road edge data; determining second point cloud data located in a lane area in the first point cloud data based on the second road edge data, and clustering the second point cloud data to obtain a candidate object; determining a target area in the lane area based on the first road edge data, and determining the candidate object in the target area as a target object to be detected. The embodiments of the present application can filter out interference data in the lane area outside, and filter out virtual targets in the lane area inside. The embodiments of the present application can filter out candidate objects outside the target area, and improve the accuracy of detecting the target object.

Description

Technical Field

[0001] This application relates to the field of autonomous driving, and more particularly to an object detection method, apparatus, electronic device, and storage medium. Background Technology

[0002] In the development of autonomous driving systems, obstacle detection has always been the most fundamental module in the perception stage. Point cloud information received by LiDAR sensors is a crucial foundation for target detection. Each LiDAR frame contains hundreds of thousands of point cloud data points, including not only information about vehicles on the road but also a detailed description of the overall road environment, including pedestrians and other obstacles.

[0003] In target detection using information-rich LiDAR point clouds, the division of road regions has always been a hot topic. This is mainly to differentiate between primary and secondary targets based on whether they are inside or outside the lane, thereby reducing the number of non-primary targets output to downstream modules, which in turn reduces the complexity of target processing in downstream modules and minimizes unnecessary overhead in target processing.

[0004] The most commonly used road region segmentation is generally based on a default region or forward-looking visual lane lines. The default region does not dynamically update the Region of Interest (ROI), making it unsuitable for various road conditions. When used for target filtering, it consistently suffers from under-filtering and over-filtering. Forward-looking visual lane lines use a three-lane model to distinguish between in-road and out-of-road target detection areas for filtering. However, when the visual three-lane model is applied to target detection filtering of LiDAR point clouds, while it can filter many secondary targets outside the road, it causes the LiDAR point cloud to lose a lot of environmental information. For example, in intersection scenes, forward-looking lane lines cannot effectively identify targets, and if the lane line distance is too short, a large number of targets within the lanes are filtered out, leading to a certain degree of missed detections. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, this application provides an object detection method, apparatus, electronic device and storage medium.

[0006] Firstly, this application provides an object detection method, including:

[0007] Obtain the first point cloud data obtained by scanning the lane area corresponding to the vehicle;

[0008] Obtain first road edge data by performing road edge detection on the first point cloud data, and perform amplification processing on the first road edge data to obtain second road edge data;

[0009] Based on the second roadside data, determine the second point cloud data located in the lane area from the first point cloud data, and obtain candidate objects by clustering based on the second point cloud data;

[0010] Based on the first roadside data, a target area is determined in the lane area, and candidate objects within the target area are identified as target objects to be detected.

[0011] Optionally, the first curb data includes multiple curb points. The first curb data is amplified to obtain second curb data, including:

[0012] Insert curb points for filling and completing the first curb data to obtain completed curb data, so that each curb point in the completed curb data corresponds to the corresponding resolution angle within the scanning field of view;

[0013] Invalid edge points for placeholders are inserted into the completed edge data to obtain second edge data, so that each resolution angle within the scanning field of view has a corresponding edge point in the second edge data.

[0014] Optionally, road edge points for filling and completion are inserted into the first road edge data to obtain completed road edge data, including:

[0015] Interpolation is performed between roadside points in the first roadside data that are spaced less than or equal to a first threshold to obtain interpolated roadside data.

[0016] The interpolated road edge data is padded between road edge points with an interval greater than the first threshold and less than the second threshold to obtain padded road edge data.

[0017] Optionally, invalid edge points used as placeholders are inserted into the completed edge data to obtain second edge data, including:

[0018] Invalid road edge points are inserted between road edge points with an interval greater than or equal to the second threshold in the completed road edge data to obtain the second road edge data.

[0019] Optionally, determining the second point cloud data located in the lane area from the first point cloud data based on the second roadside data includes:

[0020] Obtain the coordinates of each point in the first point cloud data and determine the resolution angle corresponding to each point;

[0021] Using the vehicle's location as the origin, calculate the first polar radius distance of the coordinates of the point corresponding to each resolution angle;

[0022] The first edge distance threshold corresponding to each resolution angle is determined based on the second edge data;

[0023] If any of the first extreme diameter distances is less than or equal to the corresponding first road edge distance threshold, the point is determined to be located within the lane line.

[0024] The coordinates of all points located within the lane lines are determined as the second point cloud data of the area within the lane.

[0025] Optionally, determining the first path distance threshold corresponding to each resolution angle based on the second path data includes:

[0026] In the second edge data, determine the valid edge points other than invalid edge points and the resolution angle corresponding to each valid edge point;

[0027] Calculate the first distance between the vehicle's location and each of the effective roadside points, and use it as a first distance threshold;

[0028] For invalid edge points in the second edge data, determine the resolution angle corresponding to each invalid edge point;

[0029] Based on the resolution angle, determine the corresponding default road edge point in the default road edge boundary;

[0030] Calculate the second distance between the vehicle's location and each of the default roadside points, and use it as a second distance threshold;

[0031] The first distance threshold and each of the second distance thresholds corresponding to each resolution angle are used to determine the first road edge distance threshold corresponding to each resolution angle.

[0032] Optionally, determining the target area in the lane area based on the first curb data includes:

[0033] In the first curb data, obtain the coordinates of the first curb point located to the left of the vehicle's location and farthest from the vehicle's location, and the coordinates of the second curb point located to the right of the vehicle's location and farthest from the vehicle's location.

[0034] The target area is determined based on the coordinates of the first roadside point and the second roadside point.

[0035] Optionally, determining candidate objects within the target region as target objects to be detected includes:

[0036] Obtain the coordinates of the candidate objects and determine the resolution angle corresponding to each candidate object;

[0037] Using the vehicle's location as the origin, calculate the second polar radius distance of the candidate object corresponding to each resolution angle;

[0038] Based on the target region, determine the second path distance threshold corresponding to each resolution angle;

[0039] If any second polar radius distance is less than or equal to the corresponding second path distance threshold, the candidate object is determined to be located within the target area;

[0040] Candidate objects located within the target area are identified as target objects.

[0041] Optionally, determining the target object based on candidate objects located within the target region includes:

[0042] Obtain the size features of each candidate object;

[0043] Candidate objects whose size characteristics are greater than or equal to a preset size threshold are identified as the target objects.

[0044] Secondly, this application provides an object detection device, comprising:

[0045] The first acquisition module is used to acquire the first point cloud data obtained by scanning the lane area corresponding to the vehicle;

[0046] The second acquisition module is used to acquire the first road edge data obtained by road edge detection of the first point cloud data, and to perform amplification processing on the first road edge data to obtain the second road edge data.

[0047] The first determining module is used to determine the second point cloud data located in the lane area in the first point cloud data based on the second roadside data, and to obtain candidate objects by clustering based on the second point cloud data;

[0048] The second determining module is used to determine a target area in the lane area based on the first roadside data, and to determine the candidate objects in the target area as the target objects to be detected.

[0049] Thirdly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0050] Memory, used to store computer programs;

[0051] A processor, when executing a program stored in memory, implements the object detection method described in any of the first aspects.

[0052] Fourthly, this application provides a computer-readable storage medium storing a program for an object detection method, wherein when the program for the object detection method is executed by a processor, it implements the steps of the object detection method described in any of the first aspects.

[0053] The beneficial effects of this invention are:

[0054] This application embodiment divides lane regions based on second roadside data obtained by amplifying first roadside data, resulting in second point cloud data for the lane-in-lane region. This filters out interfering data from the lane-outside region, avoiding the influence of secondary targets in the lane-outside region on the detection results. Candidate objects are clustered based on the second point cloud data of the lane-in-lane region, thus filtering out false targets in the lane-in-lane region and avoiding the influence of secondary targets in the lane-outside region on the detection results. By using the first roadside data to divide the target region, candidate objects located within the target region are identified as the target objects to be detected. This filters out candidate objects outside the target region, avoiding their influence on the detection results and improving the accuracy of target object detection. This, in turn, improves the processing efficiency of modules that rely on the detected target objects for further processing. Attached Figure Description

[0055] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 A flowchart of an object detection method provided in an embodiment of this application;

[0058] Figure 2 A schematic diagram illustrating linear interpolation of roadshow points according to resolution angle provided in this application embodiment;

[0059] Figure 3 A schematic diagram illustrating the distribution of curb points in the first curb data provided in this application embodiment;

[0060] Figure 4 This application provides a schematic diagram illustrating the distribution of roadside points in completed roadside data.

[0061] Figure 5 A flowchart of step S103 provided in an embodiment of this application;

[0062] Figure 6 A schematic diagram of an effective curb point and a default curb boundary provided for an embodiment of this application;

[0063] Figure 7A schematic diagram of second point cloud data located in a lane area provided in an embodiment of this application;

[0064] Figure 8 A schematic diagram of multiple candidate objects provided in the embodiments of this application;

[0065] Figure 9 This is a schematic diagram of the target area and target object provided in the embodiments of this application;

[0066] Figure 10 A schematic diagram of a linear interpolation method provided in an embodiment of this application;

[0067] Figure 11 This is a structural diagram of an object detection device provided in an embodiment of this application;

[0068] Figure 12 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0070] Because the default region of interest (ROI) is not dynamically updated, it cannot adapt to various road conditions, and its use for target filtering always results in under-filtering and over-filtering of targets. Forward-looking vision lane lines use a three-lane model to distinguish between in-road and out-of-road target detection ranges for target filtering. However, when the vision-based three-lane model is applied to target detection filtering of LiDAR point clouds, although it can filter many secondary targets outside the road, it causes the LiDAR point cloud to lose a lot of environmental information. For example, in intersection scenes, forward-looking lane lines cannot be effectively identified, and if the lane line distance is too short, a large number of targets within the lanes are filtered out, leading to a certain degree of missed detection. Therefore, embodiments of this application provide an object detection method, apparatus, electronic device, and storage medium.

[0071] like Figure 1 As shown in the figure, this application provides an object detection method, which may include the following steps:

[0072] Step S101: Obtain the first point cloud data obtained by scanning the lane area corresponding to the vehicle;

[0073] In this embodiment, the lane area corresponding to the vehicle refers to the road area where the vehicle is located during its travel. The first point cloud data is obtained by scanning the left, front, and right sides of the vehicle using a LiDAR scanner. In this step, the first point cloud data can be acquired from the LiDAR.

[0074] Step S102: Obtain first road edge data obtained by performing road edge detection on the first point cloud data; perform amplification processing on the first road edge data to obtain second road edge data.

[0075] In this step, road edge detection can be performed on the first point cloud data, that is, multiple road edge points in the first point cloud data can be detected to obtain the first road edge data. In one embodiment of this application, the first origin of the first point cloud data and the second origin of the first road edge point cloud are both at the center of the rear axle of the vehicle. Assuming that the vehicle's driving direction, i.e., the longitudinal direction of the vehicle, is the x-direction and the lateral direction of the vehicle is the y-direction, the relative displacement of the first origin and the second origin is in the x-direction. For subsequent use, the positions of the first origin and the second origin need to be moved to the front of the vehicle. Therefore, the origin in the first road edge data and the origin in the first point cloud data need to be moved a specified distance in the x-direction (vehicle driving direction) (i.e., the distance from the front of the vehicle to the center of the rear axle). In this way, the origin in the first road edge data can be aligned with the origin in the first point cloud data, ensuring that there is no misalignment when the point cloud is searched according to the resolution angle.

[0076] Because the density of curb points in the actual detected curb data is approximately one curb point cloud every 0.8m (the density of curb points may vary depending on the parameters), there is a one-to-one correspondence between curb points and resolution angles. Since target object detection is performed according to the resolution angle of the LiDAR, a one-to-one correspondence between curb points and resolution angles is required. Therefore, curb points need to be preprocessed, i.e., augmented, to ensure that within the LiDAR's field of view (FOV) (the LiDAR's FOV calibration parameter is 180°) at the horizontal detection angle, there are curb points as point clouds for region division at each resolution angle (0.1°).

[0077] Amplification processing refers to inserting curb points into a plurality of curb points so that the number of curb points in the second curb data obtained after the insertion reaches a specified number. This specified number can be determined based on the angular range and angular resolution of the lidar scanning system. In one embodiment of this application, the specified number = angular range / angular resolution. For example,... Figure 2 As shown, if the angle range is 180° and the angle resolution is 0.1°, then the specified quantity is 1800. After inserting the curb points, the number of curb points in the second curb data is 1800. Thus, each curb point in the second curb data can correspond one-to-one with the angle resolution.

[0078] The first and second road edge data in this application each contain multiple road edge points, that is, the road edge is represented by a set of points. This can be applied to a variety of complex road scenarios, including scenarios with road edges such as ramps and intersections. Compared with the existing technology that fits the road edge curve, there is no way to predict the road edge for parts of the actual scene where there is no road edge at a distance or near. Representing the road edge by the set of detected road edge points is more consistent with the actual detection results.

[0079] In one embodiment of this application, the first curb data includes multiple curb points. The first curb data is amplified to obtain second curb data, including:

[0080] 1) Insert curb points for filling and completing the first curb data to obtain completed curb data, so that each curb point in the completed curb data corresponds to the corresponding resolution angle within the scanning field of view;

[0081] Interpolation is performed between roadside points in the first roadside data with an interval less than or equal to a first threshold to obtain interpolated roadside data; and completion is performed between roadside points in the interpolated roadside data with an interval greater than the first threshold and less than a second threshold to obtain completed roadside data.

[0082] In other words, curb points can be inserted between curb points with intervals less than or equal to the first threshold, and curb points can be padded between curb points with intervals greater than the first threshold and less than the second threshold. Curve points can be inserted between curb points with small intervals to ensure that each curb point in the padded curb data corresponds to a corresponding resolution angle within the scanning field of view. The distribution of curb points in the first curb data can be as follows: Figure 3 As shown by the black dots, the distribution of roadside points in the completed roadside data can be illustrated as follows: Figure 4 As shown.

[0083] 2) Insert invalid edge points for placeholders into the completed edge data to obtain second edge data, so that each resolution angle within the scanning field of view has a corresponding edge point in the second edge data.

[0084] In this step, invalid curb points can be inserted between curb points with an interval greater than or equal to the second threshold in the completed curb data to obtain the second curb data.

[0085] For angles with excessive missing curb point clouds, no completion processing is performed; instead, invalid curb points are set to ensure the overall number of curbs remains unchanged. For example, between two curb points with a distance greater than 5m, it is impossible to determine whether a curb point is missing or if there is no curb in the actual scene. In this case, an invalid curb point (-∞, y, z) can be inserted as a placeholder to ensure that there are curb points in each resolution region as a threshold.

[0086] Step S103: Based on the second roadside data, determine the second point cloud data located in the lane area in the first point cloud data, and obtain candidate objects by clustering based on the second point cloud data;

[0087] In this step, the lane area can be divided into an inner lane area and an outer lane area based on multiple roadside points in the amplified second roadside data. The point cloud data located in the inner lane area of ​​the first point cloud data is determined as the second point cloud data. Clustering the second point cloud data can yield multiple candidate objects. For example, the candidate objects can be vehicles, pedestrians, or trees.

[0088] Step S104: Based on the first roadside data, determine the target area in the lane area, and identify the candidate objects in the target area as the target objects to be detected.

[0089] In this embodiment of the application, since the second curb data is obtained by amplifying the first curb data, the lane area determined based on the second curb data is larger than the lane area determined based on the first curb data.

[0090] In this step, based on multiple curb points in the first curb data, the area within the lane can be determined in the lane area to obtain the target area. Candidate objects located within the target area are then identified as target objects to be detected. For example, the target object can be a vehicle or a pedestrian.

[0091] This application embodiment divides the lane region based on the second road edge data obtained by amplifying the first road edge data, obtaining the second point cloud data of the lane region, filtering out interference data in the lane region, and avoiding the influence of secondary targets in the lane region on the detection results; candidate objects are obtained by clustering the second point cloud data of the lane region, realizing the filtering out of false targets in the lane region, and avoiding the influence of secondary targets in the lane region on the detection results; the target region is divided using the first road edge data, and then the candidate objects located in the target region are determined as the target objects to be detected, which can filter out the candidate objects outside the target region, avoid the influence of candidate objects outside the target region on the detection results, improve the accuracy of target object detection, and thus improve the processing efficiency of the subsequent modules that rely on the detected target objects for further processing. Moreover, this application only uses the point cloud information of the lidar itself, reduces module coupling, and does not need to consider time synchronization and calibration issues for multi-module matching.

[0092] In yet another embodiment of this application, as Figure 5 As shown, step S103, based on the second roadside data, determines the second point cloud data located in the lane area from the first point cloud data, including:

[0093] Step S201: Obtain the coordinates of each point in the first point cloud data and determine the resolution angle corresponding to each point;

[0094] Point cloud data includes the coordinates of multiple points. In this step, the coordinates of each point can be obtained, and the resolution angle corresponding to each point can be calculated based on its coordinates. Where point.y is the ordinate of any point and point.x is the abscissa of any point.

[0095] To map the resolution angle of each point to the curb point, we can also calculate the index of the curb point corresponding to each resolution angle.

[0096] Step S202: Using the vehicle's location as the origin, calculate the first polar radius distance of the coordinates of the point corresponding to each resolution angle;

[0097] In this embodiment of the application, the vehicle's location can refer to the position of the rear axle center of the vehicle, and the first extreme radius distance refers to the distance between each point and the origin. In this step, the first extreme radius distance...

[0098] Step S203: Determine the first path distance threshold corresponding to each resolution angle based on the second path data;

[0099] In this step, the first road edge distance threshold corresponding to each resolution angle can be calculated based on the second road edge data. The first road edge distance threshold is a judgment threshold used to divide the area inside the lane and the area outside the lane.

[0100] In this embodiment, since the second curb data includes invalid curb points inserted for placeholder purposes, other curb points besides the invalid ones can be determined as valid curb points. For valid curb points, they can be directly used as road boundaries, and a first distance between the valid curb point and the vehicle's location can be calculated as a first curb distance threshold. For invalid curb points, a default curb point corresponding to the resolution angle can be selected from the default curb boundary. That is, the default curb point is used as the road boundary, and a second distance between the default curb point and the vehicle's location is calculated as a first curb distance threshold. The first and second distances are used as the first curb distance thresholds, and the thresholds in the first curb distance threshold are arranged according to the resolution angle. In other words, the road boundary of this application includes the road boundary corresponding to the valid curb points and the road boundary corresponding to the default curb points in the scanning area where the invalid curb points are located.

[0101] This application supplements the scanning area containing invalid road edge points by using default road edge boundaries. On the one hand, it avoids directly connecting locations without road edge points (which may be intersections or locations where road edge points were not detected), thus preventing the misfiltering of potentially useful point cloud data that could constitute candidate objects and improving the accuracy of subsequent point cloud filtering. On the other hand, by using default road edge boundaries, it avoids introducing point cloud data outside the default road edge boundaries that are not useful for reference, improving the accuracy of subsequent point cloud filtering and reducing computational load.

[0102] Valid curb points and default curb boundaries can be as follows: Figure 6 As shown, Figure 6 The medium gray square grid area represents the road surface, and the side length of each square represents a distance of 10 meters. Figure 6 In the diagram, ①②③④⑤⑥⑦ represent the areas scanned by the LiDAR from first to last. In area ①, a valid roadside point is scanned, and the first distance can be calculated. In area ②, no valid roadside point is scanned, and the second distance can be calculated. In area ③, a valid roadside point is scanned, and the first distance can be calculated. In area ④, no valid roadside point is scanned, and the second distance can be calculated. In area ⑤, a valid roadside point is scanned, and the first distance can be calculated. In area ⑥, no valid roadside point is scanned, and the second distance can be calculated. In area ⑦, a valid roadside point is scanned, and the first distance can be calculated.

[0103] For example, the default curb boundary is the farthest distance in the x-direction, default_x (170m), and the maximum left and right width distance in the y-direction, default_y (20m). For instance, when the curb point is invalid in the resolution angle_tmp direction, the curb point index angle_index = 900 + (angle_tmp / 0.1°), and the corresponding default curb point in the default curb boundary in this direction is default_curb_point(x,y,z) = (default_y / tan(angle_tmp),default_y,0). The threshold (i.e., the second distance) threshold[angle_index] = sqrt(default_curb_point.x^2,default_curb_point.y^2)m. For example, when the curb point is invalid in the -45.0° direction, the curb point index angle_index = 900 + (-45.0° / 0.1°) = 450, and the range threshold in this direction is...

[0104] In one embodiment of this application, determining a first path distance threshold corresponding to each resolution angle based on the second path data includes:

[0105] 1. Determine the valid curb points (excluding invalid curb points) and the resolution angle corresponding to each valid curb point from the second curb data;

[0106] The calculation method for the resolution angle corresponding to each effective curb point is the same as the calculation method for the resolution angle in step S201, and will not be repeated here.

[0107] 2. Calculate the first distance between the vehicle's location and each of the effective roadside points, and use it as a first distance threshold;

[0108] In this embodiment of the application, the vehicle location can refer to the location of the rear axle center of the vehicle, and the first distance refers to the distance between each effective curb point and the origin. In this step, the calculation method of the first distance is the same as the calculation method of the first extreme radius distance in step S202, which will not be repeated here.

[0109] 3. For invalid edge points in the second edge data, determine the resolution angle corresponding to each invalid edge point;

[0110] The calculation method for the resolution angle corresponding to each invalid curb point is the same as the calculation method for the resolution angle in step S201, and will not be repeated here.

[0111] IV. Based on the resolution angle, determine the corresponding default curb point in the default curb boundary;

[0112] For invalid curb points, default curb points with the same resolution angle can be selected from the default curb boundary according to the resolution angle.

[0113] 5. Calculate the second distance between the vehicle's location and each of the default roadside points, and use it as a second distance threshold;

[0114] In this embodiment of the application, the vehicle location can refer to the location of the rear axle center of the vehicle, and the second distance refers to the distance between each invalid curb point and the origin. In this step, the calculation method of the second distance is the same as the calculation method of the first extreme radius distance in step S202, which will not be repeated here.

[0115] 6. Determine the first path distance threshold corresponding to each resolution angle and each of the second distance thresholds.

[0116] In this step, for each resolution angle, the first distance threshold or the second distance threshold corresponding to that resolution angle can be determined as the first path distance threshold corresponding to that resolution angle.

[0117] Step S204: If any of the first extreme diameter distances is less than or equal to the corresponding first road edge distance threshold, determine that the point is located within the lane line;

[0118] By comparing the first extreme diameter distance with the first road edge distance threshold, it can be determined whether the point corresponding to the first extreme diameter distance is located inside the lane line. If any first extreme diameter distance is less than or equal to the corresponding first road edge distance threshold, the point is determined to be located inside the lane line; if any first extreme diameter distance is greater than the corresponding first road edge distance threshold, the point is determined to be located outside the lane line.

[0119] Step S205: Determine the coordinates of all points located within the lane lines as the second point cloud data of the area within the lane.

[0120] The coordinates of all points located within the lane lines in the first point cloud data are used to determine the second point cloud data located within the lane area, such as... Figure 7 As shown; clustering the second point cloud data yields multiple candidate objects, which can be represented as follows: Figure 8 As shown, Figure 7 and Figure 8 The medium gray grid area represents the road surface. Each grid is a square, and the side length of each grid represents a distance of 10m.

[0121] This application embodiment can automatically filter out the second point cloud data located in the lane area from the first point cloud data based on the second road edge data, and filter out the interfering data located in the lane area to avoid the influence of secondary targets in the lane area on the detection results.

[0122] In another embodiment of this application, step S104, which determines a target area in the lane area based on the first curb data, includes:

[0123] Obtain the coordinates of the first curb point located to the left of the vehicle's location and farthest from the vehicle's location, and the coordinates of the second curb point located to the right of the vehicle's location and farthest from the vehicle's location, from the first curb data. For example, the coordinates with the largest x-coordinate can be selected from the coordinate sets of the curb points on the left and right sides, respectively, to serve as the coordinates of the first curb point and the second curb point.

[0124] The target area is determined based on the coordinates of the first roadside point and the second roadside point.

[0125] In this step, interpolation can be performed between the coordinates from (0, y corresponding to the left curb) to the first curb point, and the coordinates of the second curb point to the curb point at (0, y corresponding to the right curb).

[0126] In this embodiment, the interpolation process can be simplified using linear interpolation and approximate angle interpolation. Linear interpolation is as follows: Figure 10 As shown, point (x0, y0) is the known coordinates of a point along the roadside. There are k points in between that need to be interpolated to obtain their coordinates. Approximation is performed using linear interpolation based on the angle, such as... Figure 2 As shown, the number of interpolated roadside points is obtained based on the angular resolution. The approximate distance threshold does not affect the target identification. Assuming the farthest distance is 170m, the maximum error of both methods is within 0.1m and can be ignored. The linear interpolation results are shown in [reference needed]. Figure 9 For example, the curb point effect, after interpolation, also includes 1800 curb points in the target area.

[0127] In this embodiment, a closed region can be formed using the coordinates of the first path edge point and the coordinates of the second path edge point to obtain the target region, such as... Figure 9 As shown, Figure 9 The medium gray grid area represents the road surface. Each grid is a square, and the side length of each grid represents a distance of 10m.

[0128] In another embodiment of this application, step S104, which determines candidate objects within the target region as target objects to be detected, includes:

[0129] First, obtain the coordinates of the candidate objects and determine the resolution angle corresponding to each candidate object;

[0130] For example, the coordinates of a candidate object can refer to the coordinates of the center position of the candidate object, and the resolution angle corresponding to each candidate object. Where obj.x is the x-coordinate of the candidate object and obj.y is the y-coordinate of the candidate object.

[0131] To map the resolution angle of each point to the curb point, the index of the curb point corresponding to each candidate object can also be calculated.

[0132] Then, taking the vehicle's location as the origin, the second polar radius distance of the candidate object corresponding to each resolution angle is calculated;

[0133] Second polar distance

[0134] Then, based on the target area, determine the second path distance threshold corresponding to each resolution angle;

[0135] The resolution angle corresponding to each roadside point can be calculated based on the coordinates of each roadside point in the target area. Then, the distance between each roadside point in the target area and the location of the vehicle can be calculated as the second roadside distance threshold. The calculation method is the same as the calculation method of the first roadside distance threshold, and will not be repeated here.

[0136] If any second polar radius distance is less than or equal to the corresponding second path distance threshold, the candidate object is determined to be located within the target area; if any second polar radius distance is greater than the corresponding second path distance threshold, the candidate object is determined to be located outside the target area.

[0137] Finally, candidate objects located within the target region are identified as target objects. Target objects within the target region can be categorized as follows: Figure 9 As shown.

[0138] In this embodiment of the application, candidate objects located within the target area can be assigned a high confidence level, while candidate objects located outside the target area can be assigned a low confidence level. In this way, candidate objects with high confidence can be identified as target objects.

[0139] In one embodiment of this application, determining a target object based on a candidate object located within a target region includes:

[0140] Obtain the size features of each candidate object; size features can refer to information such as the length, width, and height of the candidate object.

[0141] Candidate objects whose size characteristics are greater than or equal to a preset size threshold are identified as the target objects.

[0142] This application's embodiments determine whether candidate objects are virtual targets within the lane area based on size features. For example, some ground point clouds may be detected as low-lying, small virtual targets. By using size features, especially information such as height in the z-direction, a low confidence level is set for such virtual targets within the ROI. Finally, after filtering out all low-confidence targets, all high-confidence targets are obtained as the output result. Figure 9 The obtained clustering target bounding boxes.

[0143] In another embodiment of this application, an object detection device is also provided, such as... Figure 11 As shown, it includes:

[0144] The first acquisition module 11 is used to acquire the first point cloud data obtained by scanning the lane area corresponding to the vehicle;

[0145] The second acquisition module 12 is used to acquire the first road edge data obtained by road edge detection of the first point cloud data, and to perform amplification processing on the first road edge data to obtain the second road edge data.

[0146] The first determining module 13 is used to determine the second point cloud data located in the lane area in the first point cloud data based on the second roadside data, and to obtain candidate objects by clustering based on the second point cloud data;

[0147] The second determining module 14 is used to determine a target area in the lane area based on the first roadside data, and to determine the candidate objects in the target area as the target objects to be detected.

[0148] In another embodiment of this application, an electronic device is also provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0149] Memory, used to store computer programs;

[0150] The processor, when executing a program stored in memory, implements the object detection method described in any of the foregoing method embodiments.

[0151] The electronic device provided in this invention includes a processor that executes a program stored in its memory to divide lane regions based on second road edge data obtained by amplifying first road edge data, obtaining second point cloud data for the lane region, filtering out interference data from the lane region, and avoiding the influence of secondary targets in the lane region on the detection results; clustering candidate objects based on the second point cloud data of the lane region to filter out false targets in the lane region, and avoiding the influence of secondary targets in the lane region on the detection results; using the first road edge data to divide the target region, and then determining the candidate objects located within the target region as the target objects to be detected, can filter out candidate objects outside the target region, avoid the influence of candidate objects outside the target region on the detection results, improve the accuracy of target object detection, and thus improve the processing efficiency of modules that rely on the detected target objects for further processing.

[0152] The communication bus 1140 mentioned in the above-mentioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 12 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0153] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0154] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0155] The processor 1110 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0156] In another embodiment of this application, a computer-readable storage medium is also provided, on which a program for an object detection method is stored, wherein when the program for the object detection method is executed by a processor, it implements the steps of the object detection method described in any of the foregoing method embodiments.

[0157] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0158] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An object detection method, characterized in that, include: Obtain the first point cloud data obtained by scanning the lane area corresponding to the vehicle; Obtain first road edge data by performing road edge detection on the first point cloud data, and perform amplification processing on the first road edge data to obtain second road edge data; The first curb data contains multiple curb points. The first curb data is amplified to obtain the second curb data, which includes: Insert curb points for filling and completing the first curb data to obtain completed curb data, so that each curb point in the completed curb data corresponds to the corresponding resolution angle within the scanning field of view; Invalid edge points for placeholders are inserted into the completed edge data to obtain second edge data, so that each resolution angle within the scanning field of view has a corresponding edge point in the second edge data; Based on the second roadside data, determine the second point cloud data located in the lane area from the first point cloud data, and obtain candidate objects by clustering based on the second point cloud data; Based on the first roadside data, a target area is determined in the lane area, and candidate objects within the target area are identified as target objects to be detected.

2. The object detection method according to claim 1, characterized in that, Insert curb points for filling and completion into the first curb data to obtain completed curb data, including: Interpolation is performed between roadside points in the first roadside data that are spaced less than or equal to a first threshold to obtain interpolated roadside data. The interpolated road edge data is padded between road edge points with an interval greater than the first threshold and less than the second threshold to obtain padded road edge data.

3. The object detection method according to claim 1, characterized in that, The second edge data is obtained by inserting invalid edge points for placeholders into the completed edge data, including: Invalid road edge points are inserted between road edge points with an interval greater than or equal to the second threshold in the completed road edge data to obtain the second road edge data.

4. The object detection method according to claim 1, characterized in that, Determining second point cloud data located within the lane area from the first point cloud data based on the second roadside data includes: Obtain the coordinates of each point in the first point cloud data and determine the resolution angle corresponding to each point; Using the vehicle's location as the origin, calculate the first polar radius distance of the coordinates of the point corresponding to each resolution angle; The first edge distance threshold corresponding to each resolution angle is determined based on the second edge data; If any of the first extreme diameter distances is less than or equal to the corresponding first road edge distance threshold, the point is determined to be located within the lane line. The coordinates of all points located within the lane lines are determined as the second point cloud data of the area within the lane.

5. The object detection method according to claim 4, characterized in that, Determining the first path edge distance threshold corresponding to each resolution angle based on the second path edge data includes: In the second edge data, determine the valid edge points other than invalid edge points and the resolution angle corresponding to each valid edge point; Calculate the first distance between the vehicle's location and each of the effective roadside points, and use it as a first distance threshold; For invalid edge points in the second edge data, determine the resolution angle corresponding to each invalid edge point; Based on the resolution angle, determine the corresponding default road edge point in the default road edge boundary; Calculate the second distance between the vehicle's location and each of the default roadside points, and use it as a second distance threshold; The first distance threshold and each of the second distance thresholds corresponding to each resolution angle are used to determine the first road edge distance threshold corresponding to each resolution angle.

6. The object detection method according to claim 1, characterized in that, Determining the target area within the lane area based on the first curb data includes: In the first curb data, obtain the coordinates of the first curb point located to the left of the vehicle's location and farthest from the vehicle's location, and the coordinates of the second curb point located to the right of the vehicle's location and farthest from the vehicle's location. The target area is determined based on the coordinates of the first roadside point and the second roadside point.

7. The object detection method according to claim 1, characterized in that, Determining candidate objects within the target region as target objects to be detected includes: Obtain the coordinates of the candidate objects and determine the resolution angle corresponding to each candidate object; Using the vehicle's location as the origin, calculate the second polar radius distance of the candidate object corresponding to each resolution angle; Based on the target region, determine the second path distance threshold corresponding to each resolution angle; If any second polar radius distance is less than or equal to the corresponding second path distance threshold, the candidate object is determined to be located within the target area; Candidate objects located within the target area are identified as target objects.

8. The object detection method according to claim 7, characterized in that, Candidate objects located within the target area are identified as target objects, including: Obtain the size features of each candidate object; Candidate objects whose size characteristics are greater than or equal to a preset size threshold are identified as the target objects.

9. An object detection device, characterized in that, include: The first acquisition module is used to acquire the first point cloud data obtained by scanning the lane area corresponding to the vehicle; The second acquisition module is used to acquire the first road edge data obtained by road edge detection of the first point cloud data, and to perform amplification processing on the first road edge data to obtain the second road edge data. The first curb data contains multiple curb points. The first curb data is amplified to obtain second curb data, including: inserting curb points for filling and completing the first curb data to obtain completed curb data, such that each curb point in the completed curb data corresponds to a corresponding resolution angle within the scanning field of view; and inserting invalid curb points for placeholders into the completed curb data to obtain second curb data, such that each resolution angle within the scanning field of view has a corresponding curb point in the second curb data. The first determining module is used to determine the second point cloud data located in the lane area in the first point cloud data based on the second roadside data, and to obtain candidate objects by clustering based on the second point cloud data; The second determining module is used to determine a target area in the lane area based on the first roadside data, and to determine the candidate objects in the target area as the target objects to be detected.

10. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the object detection method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for an object detection method, which, when executed by a processor, implements the steps of the object detection method according to any one of claims 1-8.

Images