Target speed estimation method, device and storage medium

Abstract

The present invention discloses a target speed estimation method, device, and storage medium. The target speed estimation method includes the following steps: determining the longitudinal position of a target relative to a vehicle; if the target is in a predetermined area in front of or behind the vehicle, selecting a first estimated speed obtained using a first estimation method as the target's estimated speed; if the target is in an area between the predetermined area in front and the predetermined area in rear, selecting a second estimated speed obtained using a second estimation method as the target's estimated speed. The present invention utilizes different sensor information acquired by current autonomous driving systems and adopts different speed estimation methods for targets in different position areas, thereby improving the accuracy of obstacle target speed estimation.

Description

Technical Field

[0001] The present invention relates to the field of unmanned driving technology, and in particular to a target speed estimation method, device and storage medium. Background Art

[0002] In autonomous driving technology, there are currently two main methods for obtaining the speed of obstacles. The first is direct measurement through millimeter-wave radar. Millimeter-wave radar can accurately obtain the relative speed of the target, but the currently commonly used millimeter-wave radar configuration has a certain observation blind spot (usually within a relatively close range to the left and right of the vehicle), so autonomous vehicles cannot obtain the speed information of targets within the blind spot. The second method is to infer the target's speed information by detecting the position change of the obstacle target through the lidar. This method has a small observation blind spot, but the inferred target speed is significantly affected by the position error measured by the lidar. Summary of the Invention

[0003] The object of the present invention is to address the technical defects existing in the prior art and provide a target speed estimation method, device and computer-readable storage medium. The target speed estimation method adopts different speed estimation methods in different position areas based on the characteristics of different sensors, thereby improving the accuracy of target speed estimation.

[0004] The technical solution adopted to achieve the purpose of the present invention is:

[0005] A target speed estimation method comprises the following steps:

[0006] Determine the position of the target relative to the vehicle in the longitudinal direction; if the target is in a predetermined area in front of the vehicle or a predetermined area behind the vehicle, select a first estimated speed obtained using the first estimation method as the estimated speed of the target; if the target is in an area between the predetermined area in front and the predetermined area behind the vehicle, select a second estimated speed obtained using the second estimation method as the estimated speed of the target.

[0007] The first estimation method obtains the first estimated speed by the following steps:

[0008] First, the tracking point of the target is selected, and then the target position information is obtained by using a laser radar or a millimeter-wave radar to obtain position and velocity information. The Kalman filter based on the kinematic model is updated in real time to obtain the first estimated velocity of the target.

[0009] The tracking point of the target is determined as follows:

[0010] If the target is located in the longitudinal area in front of or behind the vehicle, the tracking point is set to the point of the target closest to the vehicle as detected by the lidar; if the target is located in the middle area between the front and rear areas, the tracking point is set to the center point of the target as detected by the lidar.

[0011] Among them, during the tracking process, if the target crosses the area and enters the middle area from the rear area, the original tracking point is still selected for tracking in the middle area, and the tracking point selection is switched after the target crosses the middle area.

[0012] The second estimation method obtains the second estimated speed by the following steps:

[0013] Using the associated laser measurement values, calculate the positions of the upper left, upper right, lower left, and lower right corner points of the laser measurement rectangle for the target;

[0014] Using the displacement information and time difference information of each corner point, the relative speed of the four corner points and the target tracking point at time T is calculated respectively;

[0015] The speed whose value is closest to the target relative speed value at time T-1 is selected as the target speed at time T, that is, the second estimated speed.

[0016] The target speed estimation method further includes a step of estimating the speed of a new target; when a new target is established, the motion information of the new target is estimated using the following method:

[0017] When the measurement values associated with the virtual target include millimeter-wave radar measurements, the speed measurement value of the millimeter-wave radar is used as the speed estimate of the new target;

[0018] When the only measurement value associated with the virtual target is the lidar measurement, the displacement and time change of the virtual target are used to estimate the velocity of the new target.

[0019] The present invention also aims to provide a multi-sensor information fusion target speed estimation device, comprising:

[0020] a first estimating unit, configured to obtain a first estimated speed as a target estimated speed by using a first estimation method;

[0021] A second estimation unit is configured to obtain a second estimated speed as a target estimated speed using a second estimation method.

[0022] A position determination unit, configured to determine the longitudinal position of the target relative to the vehicle;

[0023] The speed selection unit is used to select the first estimated speed output by the first estimation unit as the estimated speed of the target when the target is determined to be in the predetermined area in front of the vehicle or the predetermined area behind the vehicle based on the judgment result of the position judgment unit, and to select the second estimated speed output by the second estimation unit as the estimated speed of the target when the target is determined to be in the area between the predetermined area in front of the vehicle and the predetermined area behind the vehicle.

[0024] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to implement the steps of the target speed estimation method when called by a processor.

[0025] The present invention utilizes different sensor information obtained by the current automatic driving system and adopts different speed estimation methods for targets in different position areas, thereby improving the speed estimation accuracy of obstacle targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of a target speed estimation method of the present invention;

[0027] Figure 2 Select a schematic diagram for target velocity estimation;

[0028] Figure 3 Select a schematic diagram for the target's tracking points;

[0029] Figure 4 Schematic diagram for selecting target tracking corner points and target tracking points;

[0030] Figure 5 is the principle diagram of the second estimation method;

[0031] Figure 6 Generate mechanism diagrams for new targets;

[0032] Figure 7 Schematic diagram of virtual target-associated lidar measurements;

[0033] Figure 8 This is a schematic diagram of the sensor information fusion target speed estimation device of the present invention. DETAILED DESCRIPTION

[0034] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0035] The target velocity estimation method proposed in this invention utilizes the characteristics of different sensors and adopts different velocity estimation methods in different position intervals relative to the autonomous driving vehicle (hereinafter referred to as the self-driving vehicle). This method reduces the large target velocity estimation errors caused by millimeter wave blind spots or large laser radar position errors, thereby achieving a more accurate estimation of the target velocity of perceived obstacles.

[0036] In the present invention, the target speed estimation refers to the speed of the obstacle target relative to the vehicle in the vehicle coordinate system, which can also be called the relative speed of the target.

[0037] like Figure 1 As shown, a target speed estimation method includes the following steps:

[0038] Determine the longitudinal position of a target identified as an obstacle relative to the vehicle; if the target is in a predetermined area in front of or behind the vehicle, select a first estimated speed obtained using a first estimation method as the estimated speed of the target; if the target is in an area between the predetermined area in front and the predetermined area behind the vehicle, select a second estimated speed obtained using a second estimation method as the estimated speed of the target.

[0039] For established obstacle targets, the vehicle's perception module will simultaneously calculate the speed of the current obstacle target based on two estimation methods, and then select one of the estimated speeds as the final output speed value of the obstacle target based on the target's current location area.

[0040] The selection of the above estimated speed can be found in Figure 2 As shown, the first region A is defined as the area with a longitudinal distance greater than 10m in front of the vehicle and greater than 10m behind the vehicle. The second region B is defined as the area with a longitudinal distance less than or equal to 10m but greater than or equal to -10m. Within the first region A, the target speed estimated by the first estimation method is used as the final estimated speed of obstacle target ID #1. Within the second region B, the target speed estimated by the second estimation method is used as the final estimated speed of obstacle target ID #1. The aforementioned distances of 10m and -10m are for illustrative purposes only and are not intended to be limiting. The selected distances can be used based on specific circumstances.

[0041] The aforementioned target refers to a target that has been identified as an obstacle. In the aforementioned technical solution, the first and second estimation methods each estimate the target's speed using their own estimation algorithms while the vehicle is traveling. The control system then selects and determines the target's speed based on the relationship between the target's position and the vehicle's position. By applying different speed estimation methods to targets in different location areas, the accuracy of obstacle speed estimation is improved.

[0042] The first estimation method and the second estimation method can be obtained by using the Kalman filter algorithm in the prior art or by calculating the measurement data of the laser radar.

[0043] Preferably, the first estimation method uses a Kalman filter to obtain the first estimated speed by the following method:

[0044] First, the tracking point of the target is selected according to the requirements, and then the target position information is obtained by using a laser radar or a millimeter-wave radar to obtain the position and velocity information. The Kalman filter based on the kinematic model is updated in real time to obtain the first estimated velocity of the target.

[0045] The tracking point of the target is determined as follows: Figure 2 As shown:

[0046] If the target is located in the longitudinal area in front of or behind the vehicle, the tracking point is the point closest to the vehicle as detected by the LiDAR. If the target is located in the middle area between the front and rear areas, the tracking point is set to the center point of the target as detected by the LiDAR.

[0047] Among them, during the tracking process, if the target crosses the area and enters the middle area from the rear area, the original tracking point is still selected for tracking in the middle area, and the tracking point selection is switched after the target crosses the middle area.

[0048] For details, see Figure 3 As shown, three areas are set, one area A: that is, the area with a longitudinal distance greater than 2m in front of the vehicle; one area B: that is, the area with a longitudinal distance greater than 2m behind the vehicle; one area C: the area with a longitudinal distance less than or equal to 2m in front of the vehicle and less than or equal to 2m in the rear.

[0049] If a newly generated target is located in Area A or Area B, the tracking point is the closest point to the vehicle. For vehicle ID #0, as shown in the figure, the tracking point is the black dot shown above; for vehicle ID #1, the tracking point is the black dot shown below. If a newly generated target is located in Area C, the tracking point is set to the target's center. The closest point and center point mentioned here are not the actual closest point and center point of the obstacle target, but rather the closest point and center point of the obstacle's position as detected by the LiDAR.

[0050] If the target remains within a fixed area, the target's tracking point selection method remains unchanged. If the target crosses areas during tracking, center point tracking is not used in the transition area C (-2m, 2m) to avoid switching tracking points at the boundaries. After entering transition area C, the target maintains the original tracking point selection method. However, the tracking point selection method switches when the target crosses the far edge of area C. For example, if vehicle ID #2 moves along the positive x-axis, it moves from area B to area C and then to area A. While in area B, the target's tracking point is the center point of the vehicle's front. If the target crosses the lower boundary of area C (-2m) but does not reach the upper boundary (2m), the target's tracking point remains at the center point of the vehicle's front. If the target crosses the upper boundary (2m) and enters area A, the target's tracking point switches to the center point of the vehicle's rear. If the vehicle then reenters area B, the tracking point switching boundary is at the far edge of area C (-2m).

[0051] The above-mentioned -2m, 2m is only an exemplary description and is not a limitation of the present invention.

[0052] Once the target tracking points are selected, the target's relative velocity can be determined using a Kalman filter based on an established kinematic model. Using a laser radar to obtain target position information, or a millimeter-wave radar to obtain position and velocity information, the filter is updated in real time to obtain the target's velocity information.

[0053] For established obstacle targets, when new sensor measurements are matched, the target's motion information can be updated. Current lidar detection accuracy for the center position and size of targets (larger targets such as vehicles) is not high, mainly due to factors such as the observable position of the point cloud and the target clustering effect. The visible edge position of the target obtained by lidar is relatively stable. Millimeter-wave radar generally obtains the position of the nearest point on the target, and this tracking point is inconsistent with the center point of the lidar. Therefore, in the vehicle coordinate system, by selecting the nearest point of the target measurement position as the multi-sensor fusion tracking point, the accuracy of relative velocity estimation can be improved.

[0054] By determining and switching the tracking points as described above, a good target tracking effect can be achieved, and a more accurate estimation of the target's speed can be achieved.

[0055] In the present invention, the second estimation method can use the laser radar data to calculate the second estimated speed by the following method, see Figure 4-5 As shown;

[0056] Using the associated laser measurement values, calculate the positions of the upper left, upper right, lower left, and lower right corner points of the laser measurement rectangle for the target;

[0057] Using the displacement information and time difference information of each corner point, the relative speed of the four corner points and the target tracking point at time T is calculated respectively;

[0058] The speed whose value is closest to the target relative speed value at time T-1 is selected as the target speed at time T, that is, the second estimated speed.

[0059] For details, see Figure 4-5 As shown, this method is based on the tracked target and the currently associated LiDAR measurement value, and uses the displacement of the target measurement corner points and tracking points of multiple frames to infer the relative speed of the target. When the target obtains the LiDAR measurement update value, the four corner points of the LiDAR measurement rectangle can be calculated based on the target center point position, target heading, target length and width given by the LiDAR measurement value. Due to the influence of factors such as the observable position of the obstacle and the target clustering effect of the LiDAR point cloud, the target position center point and rectangular border detected by the LiDAR are not necessarily the true center point and border position of the obstacle target. As shown in the figure, the target measurement corner points and tracking points of multiple frames are used to calculate the displacement of the target measurement corner points and tracking points of the target. When the target obtains the LiDAR measurement update value, the four corner points of the LiDAR measurement rectangle can be calculated based on the target center point position, target heading, target length and width given by the LiDAR measurement value. Due to the influence of factors such as the observable position of the obstacle and the target clustering effect of the LiDAR point cloud, the target position center point and rectangular border detected by the LiDAR are not necessarily the true center point and border position of the obstacle target. Figure 4 As shown in the figure, the gray rectangular box is the actual obstacle (such as a vehicle), and the rectangular box surrounded by four black dots is the detection result of the lidar.

[0060] The positions of the four corner points and one tracking point selected for each frame are stored in a buffer. When the valid data in the buffer meets the requirements (such as a total of 5 frames), the displacement information and time difference information of each point can be used to calculate the corresponding speed information of each point. As shown in the figure below: The buffer stores four corner point information and one target tracking point information obtained from the lidar observations of the last 5 frames. At time T, the relative speeds of the four corner points (upper left, upper right, lower left, lower right) and the target tracking point (vxrel_0, vyrel_0; vxrel_1, vyrel_1; vxrel_2, vyrel_2; vxrel_3, vyrel_3; vxrel_4, vyrel_4) are calculated respectively, and the speed calculation method for each point is calculated using the Kalman filter. After obtaining the speeds of the four corner points and one tracking point at time T, select a group of speeds from the current five groups of speeds, whose value is closest to the target relative speed value (vxrel, vyrel) at time T-1, and then use this group of speeds as the target speed at time T, such as Figure 5 shown.

[0061] In the above technical solution, the positions of the four corner points of the laser measurement rectangle for the target (the upper left, upper right, lower left, and lower right) are calculated using the associated laser measurement values. The relative speeds of the four corner points and the target tracking point at time T are then inferred based on the displacement information and time difference information. The speed closest to the target relative speed value at time T-1 is then selected as the target speed at time T, i.e., the second estimated speed, ensuring improved speed estimation accuracy under this algorithm.

[0062] It should be noted that, in the estimation of target speed, in addition to the speed estimation of the established target, it also includes the speed estimation of the newly generated or established target. Figure 6 A new target generation mechanism is shown.

[0063] The generation of a new target means that, for a certain obstacle target, after obtaining multiple associated sensor measurement values, the target is subsequently tracked as a valid obstacle target. Figure 6 This is a schematic diagram of the generation of a new target. When multiple (multi-frame) related Lidar, millimeter-wave radar or vision measurement values are obtained and the new target generation rules are met (such as the distance from the vehicle reaches a safety threshold), the new target can be established using the obtained measurement information. When establishing a new target, it is necessary to estimate the motion information of the new target.

[0064] Therefore, as an embodiment, the target speed estimation method further includes a step of estimating the speed of the new target, see Figure 7 As shown;

[0065] When a new target is established, the velocity estimation of the newly established target adopts different strategies according to the different sensor measurement categories associated with it. Specifically, the motion information of the newly established target is estimated using the following method:

[0066] When the measurement values associated with the virtual target include millimeter-wave radar measurements, the speed measurement value of the millimeter-wave radar is used as the speed estimate of the newly created target;

[0067] When the only measurement value associated with the virtual target is the lidar measurement, the displacement and time change of the virtual target are used to estimate the velocity of the newly created target.

[0068] like Figure 7 As shown in the figure, the obstacle targets detected by the lidar at time T, time T+1, and time T+2 are correlated (correlated by parameters such as target position and attributes). At time T+2, the virtual target meets the rules for generating a new target. The velocity (vx, vy) of the new target at time T+2 can be estimated according to the formula:

[0069] vx=(x2-x0) / (t2-t0), vy=(y2-y0) / (t2-t0).

[0070] From the above description, it can be seen that the present invention adopts different speed estimation methods based on the different characteristics of lidar and millimeter wave radar sensors and the different position areas where the established targets are located, thereby improving the speed estimation accuracy.

[0071] See also Figure 8 As shown, the present invention provides a multi-sensor information fusion target speed estimation device, comprising:

[0072] a first estimating unit, configured to obtain a first estimated speed as a target estimated speed by using a first estimation method;

[0073] A second estimation unit is configured to obtain a second estimated speed as a target estimated speed using a second estimation method.

[0074] A position determination unit, configured to determine the longitudinal position of the target relative to the vehicle;

[0075] The speed selection unit is used to select the first estimated speed output by the first estimation unit as the estimated speed of the target when the target is determined to be in the predetermined area in front of the vehicle or the predetermined area behind the vehicle based on the judgment result of the position judgment unit, and to select the second estimated speed output by the second estimation unit as the estimated speed of the target when the target is determined to be in the area between the predetermined area in front of the vehicle and the predetermined area behind the vehicle.

[0076] The aforementioned targets refer to established obstacle targets. For these established targets, the vehicle's perception module simultaneously calculates the speed of the current obstacle target based on two estimation methods. It then selects the speed estimated by one of the estimation methods based on the target's current location area as the final output speed value of the obstacle target.

[0077] The selection of the above estimated speed can be found in Figure 2 As shown, the first region A is defined as the area with a longitudinal distance greater than 10m in front of the vehicle and greater than 10m behind the vehicle. The second region B is defined as the area with a longitudinal distance less than or equal to 10m but greater than or equal to -10m. Within the first region A, the target speed estimated by the first estimation method is used as the final estimated speed of obstacle target ID #1. Within the second region B, the target speed estimated by the second estimation method is used as the final estimated speed of obstacle target ID #1.

[0078] The above-mentioned distances of 10m and -10m are for illustrative purposes only and are not limiting. You may choose the distances based on your specific circumstances.

[0079] The aforementioned target refers to a target that has been identified as an obstacle. In the aforementioned technical solution, the first and second estimation methods each estimate the target's speed using their own estimation algorithms while the vehicle is traveling. The control system then selects and determines the target's speed based on the relationship between the target's position and the vehicle's position. By applying different speed estimation methods to targets in different location areas, the accuracy of obstacle speed estimation is improved.

[0080] The first estimation method and the second estimation method can be obtained by using the Kalman filter algorithm in the prior art or by calculating the measurement data of the laser radar.

[0081] Preferably, the first estimation method uses a Kalman filter to obtain the first estimated speed by the following method:

[0082] First, the tracking point of the target is selected according to the requirements, and then the target position information is obtained by using a laser radar or a millimeter-wave radar to obtain the position and velocity information. The Kalman filter based on the kinematic model is updated in real time to obtain the first estimated velocity of the target.

[0083] The tracking point of the target is determined as follows: Figure 2 As shown:

[0084] If the target is located in the longitudinal area in front of or behind the vehicle, the tracking point is the point closest to the vehicle as detected by the LiDAR. If the target is located in the middle area between the front and rear areas, the tracking point is set to the center point of the target as detected by the LiDAR.

[0085] Among them, during the tracking process, if the target crosses the area and enters the middle area from the rear area, the original tracking point is still selected for tracking in the middle area, and the tracking point selection is switched after the target crosses the middle area.

[0086] Specifically, Figure 3 As shown, three areas are set, one area A: that is, the area with a longitudinal distance greater than 2m in front of the vehicle; one area B: that is, the area with a longitudinal distance greater than 2m behind the vehicle; one area C: the area with a longitudinal distance less than or equal to 2m in front of the vehicle and less than or equal to 2m in the rear.

[0087] If the newly generated target is located in area A or area B, the tracking point is the point where the obstacle target is closest to the vehicle. Figure 3For vehicle ID #0 shown in the figure, the tracking point is the point marked above; for vehicle ID #1, the tracking point is the point marked below. If the newly generated target is located in area C, the tracking point is set to the target's center. The closest point and center point mentioned here are not the actual closest point and center point of the obstacle target, but the closest point and center point of the obstacle position detected by the lidar.

[0088] If the target remains within a fixed area, the target's tracking point selection method remains unchanged. If the target crosses areas during tracking, center point tracking is not used in the transition area C (-2m, 2m) to avoid switching tracking points at the boundaries. After entering transition area C, the target maintains the original tracking point selection method. However, the tracking point selection method switches when the target crosses the far edge of area C. For example, vehicle ID #2 moves along the positive x-axis and moves from area B to area C and then to area A. While in area B, the target's tracking point is the center point of the vehicle's front. When crossing the lower boundary of area C (-2m) but not reaching the upper boundary (2m), the target's tracking point remains at the center point of the vehicle's front. When crossing the upper boundary (2m) of area C and entering area A, the target's tracking point switches to the center point of the vehicle's rear. If the vehicle then reenters area B, the tracking point switching boundary is at the far edge of area C (-2m).

[0089] The above-mentioned -2m, 2m is only an exemplary description and is not a limitation of the present invention.

[0090] Once the target tracking points are selected, the target's relative velocity can be determined using a Kalman filter based on an established kinematic model. Using a laser radar to obtain target position information, or a millimeter-wave radar to obtain position and velocity information, the filter is updated in real time to obtain the target's velocity information.

[0091] For established obstacle targets, when new sensor measurements are matched, the target's motion information can be updated. Current lidar detection accuracy for the center position and size of targets (larger targets such as vehicles) is not high, mainly due to factors such as the observable position of the point cloud and the target clustering effect. The visible edge position of the target obtained by lidar is relatively stable. Millimeter-wave radar generally obtains the position of the nearest point on the target, and this tracking point is inconsistent with the center point of the lidar. Therefore, in the vehicle coordinate system, by selecting the nearest point of the target measurement position as the multi-sensor fusion tracking point, the accuracy of relative velocity estimation can be improved.

[0092] By determining and switching the tracking points as described above, a good target tracking effect can be achieved, and a more accurate estimation of the target's speed can be achieved.

[0093] In the present invention, the second estimation method can use the laser radar data to calculate the second estimated speed by the following method, see Figure 4-5 As shown;

[0094] Using the associated laser measurement values, calculate the positions of the upper left, upper right, lower left, and lower right corner points of the laser measurement rectangle for the target;

[0095] Using the displacement information and time difference information of each corner point, the relative speed of the four corner points and the target tracking point at time T is calculated respectively;

[0096] The speed whose value is closest to the target relative speed value at time T-1 is selected as the target speed at time T, that is, the second estimated speed.

[0097] For details, see Figure 4-5 As shown, this method is based on the tracked target and the currently associated LiDAR measurement value, and uses the displacement of the target measurement corner points and tracking points of multiple frames to infer the relative speed of the target. When the target obtains the LiDAR measurement update value, the four corner points of the LiDAR measurement rectangle can be calculated based on the target center point position, target heading, target length and width given by the LiDAR measurement value. Due to the influence of factors such as the observable position of the obstacle and the target clustering effect of the LiDAR point cloud, the target position center point and rectangular border detected by the LiDAR are not necessarily the true center point and border position of the obstacle target. As shown in the figure, the target measurement corner points and tracking points of multiple frames are used to calculate the displacement of the target measurement corner points and tracking points of the target. When the target obtains the LiDAR measurement update value, the four corner points of the LiDAR measurement rectangle can be calculated based on the target center point position, target heading, target length and width given by the LiDAR measurement value. Due to the influence of factors such as the observable position of the obstacle and the target clustering effect of the LiDAR point cloud, the target position center point and rectangular border detected by the LiDAR are not necessarily the true center point and border position of the obstacle target. Figure 4 As shown in the figure, the gray rectangular box is the actual obstacle (such as a vehicle), and the rectangular box surrounded by four black dots is the detection result of the lidar.

[0098] The positions of the four corner points and one tracking point selected for each frame are stored in a buffer. When the valid data in the buffer meets the requirements (such as a total of 5 frames), the displacement information and time difference information of each point can be used to calculate the corresponding speed information of each point. As shown in the figure below: The buffer stores four corner point information and one target tracking point information obtained from the lidar observations of the last 5 frames. At time T, the relative speeds of the four corner points (upper left, upper right, lower left, lower right) and the target tracking point (vxrel_0, vyrel_0; vxrel_1, vyrel_1; vxrel_2, vyrel_2; vxrel_3, vyrel_3; vxrel_4, vyrel_4) are calculated respectively by displacement, and the speed calculation method for each point is calculated using the Kalman filter. After obtaining the speeds of the four corner points and one tracking point at time T, select a group of speeds from the current five groups of speeds, whose value is closest to the target relative speed value (vxrel, vyrel) at time T-1, and then use this group of speeds as the target speed at time T, such as Figure 5 shown.

[0099] In the above technical solution, the positions of the four corner points of the laser measurement rectangle for the target (the upper left, upper right, lower left, and lower right) are calculated using the associated laser measurement values. The relative speeds of the four corner points and the target tracking point at time T are then inferred based on the displacement information and time difference information. The speed closest to the target relative speed value at time T-1 is then selected as the target speed at time T, i.e., the second estimated speed, ensuring improved speed estimation accuracy under this algorithm.

[0100] It should be noted that, in the estimation of target speed, in addition to the speed estimation of the established target, it also includes the speed estimation of the newly generated or established target. Figure 6 A new target generation mechanism is shown.

[0101] The generation of a new target means that, for a certain obstacle target, after obtaining multiple associated sensor measurement values, the target is subsequently tracked as a valid obstacle target. Figure 6 This is a schematic diagram of the generation of a new target. When multiple (multi-frame) related Lidar, millimeter-wave radar or vision measurement values are obtained and the new target generation rules are met (such as the distance from the vehicle reaches a safety threshold), the new target can be established using the obtained measurement information. When establishing a new target, it is necessary to estimate the motion information of the new target.

[0102] Therefore, as an embodiment, the target speed estimation method further includes a step of estimating the speed of the new target, see Figure 7 As shown;

[0103] When a new target is established, the velocity estimation of the newly established target adopts different strategies according to the different sensor measurement categories associated with it. Specifically, the motion information of the new target is estimated using the following method:

[0104] When the measurement values associated with the virtual target include millimeter-wave radar measurements, the speed measurement value of the millimeter-wave radar is used as the speed estimate of the new target;

[0105] When the only measurement value associated with the virtual target is the lidar measurement, the displacement and time change of the virtual target are used to estimate the velocity of the new target.

[0106] like Figure 7 As shown in the figure, the obstacle targets detected by the lidar at time T, time T+1, and time T+2 are correlated (correlated by parameters such as target position and attributes). At time T+2, the virtual target meets the rules for generating a new target. The velocity (vx, vy) of the new target at time T+2 can be estimated according to the formula:

[0107] vx=(x2-x0) / (t2-t0), vy=(y2-y0) / (t2-t0).

[0108] From the above description, it can be seen that the present invention adopts different speed estimation methods based on the different characteristics of lidar and millimeter wave radar sensors and the different position areas where the established targets are located, thereby improving the speed estimation accuracy.

[0109] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to implement the steps of the target speed estimation method when called by a processor.

[0110] The present invention utilizes different sensor information obtained by the current automatic driving system and adopts different speed estimation methods for targets in different position areas, thereby improving the speed estimation accuracy of obstacle targets.

[0111] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A target speed estimation method, characterized in that: The following steps are involved: Determine the target's longitudinal position relative to the vehicle; If the target is in a predetermined area in front of or behind the vehicle, a first estimated speed obtained using the first estimation method is selected as the estimated speed of the target; if the target is in an area between the predetermined area in front and the predetermined area behind the vehicle, a second estimated speed obtained using the second estimation method is selected as the estimated speed of the target; The first estimation method obtains the first estimated speed by the following steps: First, the target tracking point is selected, and then the target position information is obtained using a laser radar or a millimeter-wave radar to obtain position and velocity information. The Kalman filter based on the kinematic model is updated in real time to obtain the first estimated velocity of the target. The tracking points of the target are determined as follows: If the target is located in the longitudinal area in front of or behind the vehicle, the tracking point is set to the point closest to the vehicle as detected by the LiDAR. If the target is located in the middle area between the front and rear areas, the tracking point is set to the center point of the target as detected by the LiDAR. The second estimation method obtains a second estimated speed by the following steps: Using the associated laser measurement values, calculate the positions of the upper left, upper right, lower left, and lower right corner points of the laser measurement rectangle for the target; Using the displacement information and time difference information of each corner point, the relative speed of the four corner points and the target tracking point at time T is calculated respectively; The speed whose value is closest to the target relative speed value at time T-1 is selected as the target speed at time T, that is, the second estimated speed.

2. The target speed estimation method according to claim 1, characterized in that: During the tracking process, if the target crosses the area and enters the middle area from the rear area, the original tracking point will still be selected for tracking in the middle area. After the target crosses the middle area, the tracking point selection will be switched.

3. The target speed estimation method according to claim 1 or 2, characterized in that: The process also includes the step of estimating the velocity of the new target. When a new target is established, the motion information of the new target is estimated using the following method: When the measurement values associated with the virtual target include millimeter-wave radar measurements, the speed measurement value of the millimeter-wave radar is used as the speed estimate of the new target; When the only measurement value associated with the virtual target is the lidar measurement, the displacement and time change of the virtual target are used to estimate the velocity of the new target.

4. A multi-sensor information fusion target speed estimation device, characterized in that: include: a first estimating unit, configured to obtain a first estimated speed as a target estimated speed by using a first estimation method; A second estimation unit is configured to obtain a second estimated speed as a target estimated speed using a second estimation method. A position determination unit, configured to determine the longitudinal position of the target relative to the vehicle; a speed selection unit configured to select, based on a determination result of the position determination unit, the first estimated speed output by the first estimation unit as the estimated speed of the target when the target is determined to be in a predetermined area in front of or behind the vehicle, and to select the second estimated speed output by the second estimation unit as the estimated speed of the target when the target is determined to be in an area between the predetermined area in front and the predetermined area behind the vehicle; The first estimation method obtains a first estimated speed by the following method: First, the target tracking point is selected, and then the target position information is obtained using a laser radar or a millimeter-wave radar to obtain position and velocity information. The Kalman filter based on the kinematic model is updated in real time to obtain the first estimated velocity of the target. If the target is located in the longitudinal area in front of or behind the vehicle, the tracking point is set to the point closest to the vehicle as detected by the LiDAR. If the target is located in the middle area between the front and rear areas, the tracking point is set to the center point of the target as detected by the LiDAR. The second estimation method obtains a second estimated speed by the following method; Using the associated laser measurement values, calculate the positions of the upper left, upper right, lower left, and lower right corner points of the laser measurement rectangle for the target; Using the displacement information and time difference information of each corner point, the relative speed of the four corner points and the target tracking point at time T is calculated respectively; The speed whose value is closest to the target relative speed value at time T-1 is selected as the target speed at time T, that is, the second estimated speed.

5. The multi-sensor information fusion target speed estimation device according to claim 4, characterized in that: During the tracking process, if the target crosses the area and enters the middle area from the rear area, the original tracking point will still be selected for tracking in the middle area. After the target crosses the middle area, the tracking point selection will be switched.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the target speed estimation method according to any one of claims 1 to 3 when called by a processor.

Images