An object detection method, system and computer-readable storage medium for autonomous driving

Abstract

The present invention provides an object detection method, system and computer-readable storage medium for autonomous driving. The method associates and fuses the detected object information with the passable area, and outputs the object results that can be associated or the objects whose occurrence times exceed a set threshold, and does not output the results that cannot be associated or whose occurrence times do not reach the set value. By fusing and judging the object information and the passable area, the present invention can improve the performance of object detection, effectively improve the deficiencies of the object detection module, reduce the false detection rate of object detection, and provide an effective judgment basis for the planning and decision-making of the driverless system.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and particularly relates to target detection technology. Background Art

[0002] Autonomous driving cannot do without the mutual cooperation of perception, control, and execution. Among them, sensors are the hardware basis for perceiving the surrounding environment. Sensors such as cameras and radars can obtain information such as images, distances, and speeds. At present, the vast majority of the causes of safety accidents in autonomous driving occur in this important link of sensors. Since each sensor has its own insurmountable defects, simply increasing the quantity cannot solve the actual problems. Integrating the information collected by different sensors is the real solution. Sensor fusion combines the data and information obtained by multiple sensors for comprehensive analysis to more accurately and reliably detect the surrounding environment, such as detecting targets such as vehicles and pedestrians and detecting passable areas, improving the correctness of system decisions.

[0003] In an unmanned driving system, there is a problem of false detection at the target level, such as detecting virtual targets that do not exist in the actual environment. The false detection defect of target detection has a significant negative impact on the planning and decision-making system, such as causing false braking, etc., leading to serious safety hazards.

[0004] Patent document CN111695619A discloses a method for multi-sensor target fusion. In this method, the original radar signals collected by multiple forward millimeter-wave radars and the original camera signals collected by a camera are obtained, the radar matching signals and camera matching signals that match the target are screened out, and the radar matching signals and camera matching signals are fused to output target fusion information. The use of the fusion of different sensors in this method reduces the false detection rate and the missed detection rate to a certain extent. However, effective non-homogeneous information is not used to further verify the target information, and problems such as target false detection may still exist. Summary of the Invention

[0005] In the process of multi-sensor fusion, in addition to fusing the homogeneous signals of different sensors, heterogeneous signals can also be fused for mutual verification. For example, the target information of different sensors is fused and verified with the passable area information, or the target information and passable area information after the fusion of different sensors are fused and verified. Therefore, aiming at the deficiencies of the prior art, the present invention aims to implement a target detection method, system, and computer-readable storage medium for autonomous driving, which fuse the passable area with the target information to improve the performance of target detection and reduce the false detection rate of target detection.

[0006] The technical solution of the present invention is as follows:

[0007] An object detection method for autonomous driving, which associates and fuses the detected object information with the passable area, and outputs the object results that can be associated or the objects whose occurrence times exceed the set threshold, and does not output the results that cannot be associated or whose occurrence times do not reach the set value.

[0008] In the perception system of driverless vehicles, the object detection module can output object information of the surrounding environment, such as type, position, distance, speed, etc.; the passable area module in the perception system can output the passable area (i.e., freespace) in the surrounding environment. The passable area refers to the area where the vehicle can drive, and there are no objects in the passable area. This method further determines the presence or absence of an object by fusing the passable area information with the object information to improve the object detection rate.

[0009] Specifically, the method of the present invention includes the following steps:

[0010] Step 1, data reading:

[0011] Read the detected object data, including information such as the type, position, speed, and occurrence times of the object;

[0012] Read the key points of the passable area module, and the connection of the key points forms the envelope of the passable area.

[0013] Step 2, data fusion:

[0014] Calculate the range of the association area according to the position of the object, the error value of the sensor, etc., and search for the key points of the feasible driving area within the association area. If the key points of the passable area are found within the association area, and the type and speed attributes of the key points are consistent with the type and speed attributes of the object, then the object exists; if no key points are found within the association area, then judge the occurrence times of the object. If it exceeds the set times, it is considered that the object exists and output the object; if it does not exceed the set times, do not output the object and enter the next processing cycle.

[0015] Step 3, data output: Output the object information confirmed for output after data fusion processing.

[0016] On the other hand, the present invention also proposes an object detection system for autonomous driving, which includes a processor and a memory. A computer program is stored on the memory, and when the computer program is executed by the processor, the object detection method for autonomous driving as described in the above technical solution is implemented.

[0017] On the other hand, the present invention also proposes a computer-readable storage medium, on which a computer program is stored. The computer program is characterized in that when the computer program is executed by a processor, the object detection method for autonomous driving as described in the above technical solution is implemented.

[0018] The present invention can optionally use current mainstream camera sensors, millimeter-wave radar sensors, etc., and during the process of autonomous driving, it can output target information and passable areas in real time after processing. By fusing and judging the target information and passable areas, the present invention improves the performance of target detection, can effectively improve the deficiencies of the target detection module, reduce the false detection rate of target detection, and provide an effective judgment basis for the planning and decision-making of the driverless system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a logic flowchart of the fusion of the passable area and target information. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The following further describes the system for fusing the passable area and target information with reference to the drawings.

[0021] The present invention focuses on the associative fusion of target information and passable areas, and outputs the results that can be associated or the targets whose occurrence times exceed the set threshold, and does not output the results that cannot be associated or whose occurrence times do not reach the set value.

[0022] See Figure 1 , the fusion of the passable area and target information includes three parts: data reading, data fusion processing, and data output.

[0023] 1. Data reading:

[0024] The target detection system for the fusion of the passable area and target information is composed of common sensors for driverless driving, including cameras and millimeter-wave radars.

[0025] After the system for fusing the passable area and target information is started, first, the information processed by the target processing module and the passable area processing module is collected, including the target information output by the target processing module, including the position coordinates, occurrence times, types, speeds, etc. of the targets, and the key point information output by the passable area processing module, including position coordinate values, types, speeds, etc., and the connection lines thereof form the envelope of the passable area.

[0026] 2. Data fusion processing:

[0027] Calculate the range of the associated area according to the position of the target, the error value of the sensor, etc., find the key points of the passable area within the associated area, and further process the target information according to the association result. If within the associated area, the key points of the passable area are found and the type and speed attributes of the key points are consistent with the type and speed attributes of the target, then the target exists; if no key points are found within the associated area, then judge the occurrence times. If it exceeds the set value, it is considered that the target exists; if it does not exceed the set number of times, the target is not output and enters the next processing cycle.

[0028] In this embodiment, the specific processing is as follows:

[0029] Calculate the associated region Ass_Region according to the position coordinates (x_target, y_target) of the target and the error information of the sensor processing module. Ass_Region is set as a circular region, and the representation is as follows:

[0030] Ass_Region_2 = (x0, y0, R); (1)

[0031] Among them, (x0, y0) and R are the origin coordinates and radius of the circular region respectively.

[0032] The calculation formulas for the origin coordinates and radius of the associated region Ass_Region are as follows:

[0033] x0 = x_target; y0 = y_target; (2)

[0034] R = err_obj_process + err_freespace_process + compensation; (3)

[0035] Among them, compensation is the compensation value, (x_target, y_target) is the position coordinate value of the target, and the coordinate position is at the center of the target. err_obj_process is the position error value of the target processing module, and err_freespace_process is the key point error value of the passable module. err_obj_process and err_freespace_process come from the test statistical results of the target processing module and the passable area processing module. If the above error values cannot be obtained, the approximation can be obtained by the following formula:

[0036] err_obj_process = Max(err_obj_radar, err_obj_camera) (4)

[0037] err_freespace_process = Max(err_freespace_radar, err_freespace_camera) (5)

[0038] Among them, err_obj_radar is the position error value of the target detection by the millimeter-wave radar, err_obj_camera is the position error value of the target detection by the camera, err_freespace_radar is the error value of the key points in the passable area of the millimeter-wave radar, and err_freespace_camera is the error value of the key points in the passable area of the camera. Max() is the function to take the maximum value.

[0039] Next, determine whether the key points output by the passable area module are within the associated area Ass_Region. The calculation formula is as follows:

[0040] dist = sqrt((xi - x0)*(xi - x0)+(yi - y0)*(yi - y0));

[0041] Among them, (xi, yi) are the coordinates of the key points in the passable area, and sqrt() is the square root calculation function.

[0042] If dist > R, then the key point is not within the associated area Ass_Region, otherwise the key point is within the associated area.

[0043] If there are key points within the associated area Ass_Region, continue to determine whether other attributes are consistent. The attributes include but are not limited to:

[0044] The target type type_target and the type of the key point type_freespace. For example, the types are the same for cars, trucks, pedestrians, etc.

[0045] The target speed vel_target and the key point speed vel_freespace. If the speed satisfies:

[0046] vel_target > vel_freespace * (1 - vel_para)

[0047] And vel_target < vel_freespace * (1 + vel_para)

[0048] Among them, vel_para is an adjustment parameter, which is a positive number, such as 0.15.

[0049] If the key point is within the associated area Ass_Region and the attributes of the target and the key point are consistent, then confirm the output target; if there are no key points within the associated area, then proceed to the next judgment.

[0050] Next, compare the number of occurrences of the target Obj_Age with the set threshold Age_Threshold. Among them, Age_Threshold is a positive integer, for example, the value is 7.

[0051] If the number of occurrences of the target Obj_Age is not less than the set threshold Age_Threshold, the target information is output, otherwise the target is not output.

[0052] 3. Data output: Output the target information confirmed after data fusion processing.

[0053] The specific implementation manners of the present invention described above do not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and deformations made according to the technical concept of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. An object detection method for autonomous driving, characterized in that, The method is to associate and fuse the detected target information with the passable area, and output the target results that can be associated or the targets whose occurrence times exceed the set threshold, and do not output the results that cannot be associated or whose occurrence times do not reach the set value; it includes the following steps: Step 1, Data reading: Read the detected target data, including the type, location, speed, and occurrence times information of the target; Read the key points of the passable area module, and the connection of the key points forms the envelope of the passable area; Step 2, Data fusion: Calculate the range of the association area according to the location of the target and the error value of the sensor, and find the key points of the passable area within the association area. If the key points of the passable area are found within the association area, and the type and speed attributes of the key points are the same as those of the target, then the target exists; if no key points are found within the association area, then judge the occurrence times of the target. If it exceeds the set times, then the target is considered to exist and the target is output; if it does not exceed the set times, then the target is not output and enters the next processing cycle; The range of the association area is calculated as follows: According to the position coordinates (x_target, y_target) of the target and the error information of the sensor processing module, calculate the association area Ass_Region, and Ass_Region is set as a circular area, as follows: Ass_Region_2 = (x0, y0, R) (1) Among them, (x0, y0) and R are the origin coordinates and radius of the circular area respectively; The calculation formulas for the origin coordinates and radius of the association area Ass_Region are as follows: x0 = x_target; y0 = y_target (2) R = err_obj_process + err_freespace_process + compensation (3) Among them, compensation is the compensation value, (x_target, y_target) is the position coordinate value of the target, the coordinate position is at the center of the target, err_obj_process is the position error value of the target processing module, err_freespace_process is the key point error value of the passable module, and err_obj_process and err_freespace_process come from the test statistical results of the target processing module and the passable area processing module; If the above error values cannot be obtained, the approximation is obtained using the following formula: err_obj_process = Max(err_obj_radar, err_obj_camera) (4) err_freespace_process = Max(err_freespace_radar, err_freespace_camera) (5) Among them, err_obj_radar is the position error value of the target detection of the millimeter-wave radar, err_obj_camera is the position error value of the target detection of the camera, err_freespace_radar is the error value of the key points of the passable area of the millimeter-wave radar, err_freespace_camera is the error value of the key points of the passable area of the camera, and Max() is the maximum value function; Step 3, data output: Output the target information confirmed after data fusion processing.

2. The object detection method for autonomous driving according to claim 1, wherein Judge whether the key points output by the passable area module are within the associated area Ass_Region, and the calculation formula is as follows: dist = sqrt((xi - x0)*(xi - x0)+(yi - y0)*(yi - y0)) Among them, (xi, yi) are the coordinates of the key points of the passable area, and sqrt() is the square root calculation function; If dist > R, the key point is not within the associated area Ass_Region, otherwise the key point is within the associated area.

3. The object detection method for autonomous driving according to claim 1, characterized in that, The speed attributes of the said target are consistent when the speed satisfies: vel_target > vel_freespace * (1 - vel_para) And vel_target < vel_freespace * (1 + vel_para) Among them, vel_target is the target speed, vel_freespace is the key point speed, and vel_para is the adjustment parameter, which is a positive number.

4. An object detection system for autonomous driving, characterized in that, It includes a processor and a memory, and a computer program is stored on the memory. When the computer program is executed by the processor, it realizes the target detection method for autonomous driving as described in any one of claims 1-3. When the computer program is executed by the processor, it realizes the target detection method for autonomous driving as described in any one of claims 1-3.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it realizes the target detection method for autonomous driving as described in any one of claims 1-3.

Images