Automatic driving parking real-time self-adaptive looking-around parking space tracking method and vehicle

Through the real-time adaptive surround view parking space tracking method of autonomous driving parking, high, medium and low-precision parking space tracking is carried out using visual surround view perception results and conversion matrix updates, which solves the problem of poor robustness caused by abnormal vehicle and sensor parameters and improves automatic Safety and real-time driving.

CN118991745APending Publication Date: 2024-11-22SCI & TECH CO LTD HEFEI INTELLIGENT VEHICLE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411250802.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-06
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the case of abnormal vehicle and sensor parameters, existing autonomous driving vehicles have poor parking space tracking, which cannot provide a comprehensive real-time safe driving reference, especially the problems of poor real-time ultrasonic radar perception and poor visual surround view splicing accuracy.

Method used

The real-time adaptive surround view parking space tracking method is adopted for automatic parking parking, and the point cloud processing is performed through the verification and conversion matrix checksum of the visual surround view perception result, and parking space tracking is carried out in the high, medium and low precision ranges, and the successful tracking results are merged. Provide the latest parking space information to ensure robustness and real-time.

Benefits of technology

Improves the robustness of vehicle parking space tracking under abnormal parameters, enhances the safety of autonomous driving parking, solves the problems of poor real-time perception of ultrasonic radar and poor visual surround view splicing accuracy, and provides continuous and comprehensiveness for autonomous driving. parking space reference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118991745A_ABST
    Figure CN118991745A_ABST
Patent Text Reader

Abstract

The invention discloses a real-time self-adaptive surround-view parking space tracking method for automatic driving parking and a vehicle. The method comprises the following steps: S1, checking a visual surround-view perception result and updating a conversion matrix based on vehicle driving information; s2, for the visual perception parking space, the historical tracking parking space and the conversion matrix which are verified successfully, performing point cloudization on the parking space and performing alignment processing to the current moment; s3, performing parking space tracking with high, medium and low precisions on the parking space angular points in sequence, merging the successfully tracked parking spaces to obtain a tracked available parking space, otherwise, entering parking space tracking in a next precision range, and newly building a parking space after all tracking fails; and S4, taking the current parking space tracking result as a historical tracking parking space at the next moment, repeatedly executing the steps S1-S3, and continuously obtaining the latest parking space tracking result. According to the invention, the problem of poor parking space tracking robustness under the condition of abnormal parameters of the vehicle and the sensor is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and in particular to a real-time adaptive surround-view parking space tracking method for autonomous driving parking and a vehicle using the method. Background Art

[0002] Parking space tracking plays a vital role in the parking process of autonomous vehicles. It can provide real-time parking space information for vehicles. The defined parking space features can provide vehicles with a safe parking range reference, optimize local vehicle planning and vehicle control, and thus promote the realization of autonomous driving. Therefore, parking space tracking calculation is a key technology in the field of autonomous driving.

[0003] At present, the surround-view parking tracking algorithms deployed on the vehicle side are mostly a priori results, which are closely related to the vehicle body parameters and sensor parameters. When the parameters of the vehicle and sensor change after long-term use, the adaptability is poor and the robustness is limited. It cannot provide a good reference guarantee for all-round parking and real-time safe driving of vehicles.

[0004] For example, the spatial parking space perception based on ultrasonic radar is simple to calculate and has low cost, but is limited by the reflection range and requires the vehicle to pass completely before parking in the space. The real-time performance is poor, and the influence of the sensor installation angle and other factors can easily lead to abnormal calculation of the parking space size, and it is not possible to track the available parking spaces well.

[0005] For example, parking space tracking based on visual surround view and other perception information has a good ability to recognize available parking spaces within the image range, but is limited by the stability of surround view stitching and perception accuracy (for example, the farther the surround view stitching is from the vehicle body, the worse the position perception accuracy of the parking space), and its robustness is limited.

[0006] For example, parking space tracking based on traditional algorithms mostly works in the case where sensor parameters remain unchanged. It has poor adaptability to scenarios where the installation angle changes due to abnormal situations such as collisions between vehicles or sensors, and it is difficult to adaptively generate robust parking space tracking results.

[0007] With the development of autonomous driving technology and the continuous breakthroughs in automatic parking technology, the stable tracking and updating of parking space data through adaptive algorithms to obtain more robust parking space information will provide a comprehensive safety reference guarantee for autonomous driving parking. Summary of the invention

[0008] The object of the present invention is to provide a real-time adaptive surround-view parking space tracking method, a computer program product and a vehicle for autonomous vehicle parking, so as to solve the problem of poor robustness of parking space tracking when vehicle and sensor parameters are abnormal.

[0009] To this end, the present invention provides a real-time adaptive surround view parking space tracking method for automatic driving parking, comprising the following steps: S1, verifying the visual surround view perception result and updating the conversion matrix based on the vehicle driving information; S2, for the successfully verified visual perception parking space, the historical tracking parking space and the conversion matrix, the parking space is point clouded and aligned to the current moment; S3, the parking space corner points are tracked in three different precisions of high, medium and low in turn, the successfully tracked parking spaces are merged to obtain the tracked parking spaces, otherwise, the next parking space tracking is entered into the next precision range, and a new parking space is created for all tracking failures; S4, the current parking space tracking result is used as the historical tracking parking space at the next moment, and the above steps S1-S3 are repeatedly executed to continuously obtain the latest parking space tracking results.

[0010] Compared with the prior art, the automatic driving parking real-time adaptive surround-view parking space tracking method of the present invention has the following advantages:

[0011] (1) Providing an adaptive parking space tracking method can effectively solve the problem of poor robustness of parking space tracking when vehicle and sensor parameters are abnormal, thereby improving the safety of autonomous parking;

[0012] (2) Providing a lightweight online real-time parking space tracking method that uses surround view perception results as input can effectively solve the problem of poor real-time performance of perception tracking results such as ultrasonic radar;

[0013] (3) Providing a solution to the problem that the perception accuracy of surround view stitching results gradually deteriorates from near to far, which can effectively solve the tracking anomalies and safety issues of surround view perception parking spaces during parking, and provide a continuous and comprehensive reference basis for autonomous driving.

[0014] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0016] Figure 1 is a flow chart of the real-time adaptive surround-view parking space tracking method for automatic driving parking of the present invention;

[0017] Figure 2 It is a schematic diagram of the visual surround parking space perception result defined in the real-time adaptive surround parking space tracking method for autonomous driving parking of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0019] like Figure 1 As shown, the real-time adaptive surround-view parking space tracking method for automatic driving parking of the present invention comprises the following steps:

[0020] S1, source data initialization, update conversion matrix based on vehicle driving information, and verify based on visual surround perception results;

[0021] The visual surround parking space perception result is defined as the objective existing lined parking space estimation under the current state of the vehicle, which mainly includes the parking space attribute ID (Identification) and the coordinates (x, y, z) of the four corner points of the parking space under the vehicle body system corresponding to the ID. The perceived parking space corner points are arranged counterclockwise in a rectangle, from 1 to 4, and the upper right corner of the parking space relative to the vehicle body is ranked 1, such as Figure 2 shown.

[0022] S2, based on the parking spaces verified successfully by visual perception in step S1, the historically tracked parking spaces and the conversion matrix, point cloud the parking spaces and align them to the current moment;

[0023] S3, tracking parking spaces in three different precision ranges of high, medium and low is performed in sequence, the successfully tracked parking spaces are merged to obtain the tracked available parking spaces, and new parking spaces are created for the failed tracking;

[0024] S4. The current parking space tracking result is used as the historical tracking parking space at the next moment, and steps S1-S3 are repeated to continuously update and obtain the latest parking space tracking result.

[0025] The steps of updating the conversion matrix based on the vehicle driving information in step S1 are as follows.

[0026] 1) Determine the validity of vehicle driving information, mainly including whether the input time and parameter range are consistent;

[0027] The validity of the vehicle driving information is determined, and the results with zero or abnormal values ​​for time, vehicle speed and yaw rate are eliminated.

[0028] 2) Calculate the yaw angle, coordinates and transformation matrix of the vehicle based on the starting position through the vehicle speed and yaw rate (yawrate) in the vehicle driving information;

[0029] Basic formula for angle: Where t is the current time (the starting time is set to 0 by default);

[0030] Distance calculation formula: Where speed tis the speed at time t, Δt is the time difference between time t and time t-1;

[0031] Coordinate update: Where (x t ,y t ) is the vehicle coordinate at time t (the default height coordinate z is not updated in the plane);

[0032] Transformation Matrix: Where T t is the current position of the vehicle in the first frame coordinate system [x t ,y t ,z t ] T , R t is the rotation matrix calculated based on the yaw angle

[0033] The steps of verifying based on the visual surround perception results in step S1 are as follows:

[0034] 1) Determine the validity of the parking space in the perception result, mainly including whether the parking space coordinate range is in compliance;

[0035] The validity of the visual surround parking space perception results is determined, and the results with zero or abnormal values ​​of parking space coordinates are eliminated.

[0036] In step S2, the steps of converting the parking spaces into point clouds and aligning them to the current moment based on the successfully verified parking spaces, historical parking spaces and the conversion matrix are as follows:

[0037] 1) Track the current and historical parking spot conversion based on the vehicle body transformation matrix.

[0038] Based on the transformation matrix The parking space coordinate information at multiple moments is converted to the vehicle body coordinate system at the current moment.

[0039] Conversion formula: Where P w is the coordinate in the first frame coordinate system (x w ,y w ,z w ), P c is the coordinate of the vehicle body coordinate system at a certain moment (x c ,y c ,z c ).

[0040] The steps of high-precision parking space tracking / merging in step S3 are as follows:

[0041] 1) Perform iterative close point (ICP) registration on the parking lot obtained in step S1;

[0042] The high-precision registration error threshold H is set to be less than 0.1m during implementation, that is, centimeter-level perception accuracy. The registration formula is as follows:

[0043] formula:

[0044] The original point cloud is p′ i =[x′ i ,y′ i ,z′ i ] coordinate point set (historical tracking parking space corner point coordinates), the target point cloud is p i =[x i ,y i ,z i ] is the coordinate point set (i.e. the coordinates of the corner points of the surround perception parking space at the current moment), i∈(1,n), n is the number of point sets, R and t are the rotation matrix and displacement vector based on the initial vehicle posture. The smaller the residual value J, the better the matching.

[0045] In specific implementation, for the same free parking space, the vehicle visually perceives the parking space at the current time t and the previous time t-1 (the parameters include the attribute ID perceived twice and the four corner points (x, y, z) of the vehicle body at the corresponding time), and the registration process is as follows:

[0046] First, the RT matrix is ​​used to convert the corner coordinates of the parking space at time t-1 to the body system at time t (see the above formula for the conversion formula), and then the ICP alignment is performed under the body system at time t.

[0047] 2) If the registration result meets the accuracy requirements, the parking spaces are merged and output.

[0048] If the registration error is less than the threshold H, the registration is considered successful, and the id and coordinates are merged (the default id is taken at time t-1, and the coordinates are taken at time t). Otherwise, the registration is considered failed, and the next step of medium-precision parking space tracking is continued.

[0049] The steps of merging the medium-precision parking space tracking in step S3 are as follows:

[0050] 1) Perform IOU (Intersection Over Union) registration on the parking spaces obtained in step S1;

[0051] Registration formula: A is the plane surrounded by the four corner points of the parking space to be matched at the current moment, and B is the plane surrounded by the four corner points of the parking space to be matched in the historical tracking.

[0052] In the specific implementation, the coordinates of the four corner points of the parking space A plane are P 1 ,P2 ,P 3 ,P 4 , for the coordinates of the four corner points of the parking space B plane, take P′ 1 ,P′ 2 ,P′ 3 ,P′ 4 , the corner point coordinates only take the x and y coordinates, and the z coordinate is set to 0 and does not participate in the calculation.

[0053] In order to solve the problem of non-vertical angle between the parking space and the vehicle body, the local coordinate system will be established with the two vertical sides of the parking space to be matched at the current moment, so as to output higher-precision coordinates. Get the local coordinate system transformation matrix All the corner points of the parking spaces to be matched are based on RT i All are converted to the local coordinate system for calculation, where T i The coordinates of the corner point of the parking space with the current perception result id i of the vehicle are selected, and the P in the second position is selected. 2 corner point based on P 2 Calculating the yaw angle R i is the rotation matrix calculated based on the yaw angle

[0054] For the parking space planes A and B after conversion, the specific calculation steps are as follows: The coordinates of the upper left corner are:

[0055]

[0056] The coordinates of the lower right corner of the intersection are:

[0057]

[0058] The area of ​​the intersection is calculated as:

[0059] intersection=|max(x 2 -x 1 ,0)×max(y 1 -y 2 ,0)|

[0060] The area of ​​parking space planes A and B is:

[0061]

[0062] Calculate the area of ​​the compared part:

[0063] union=S A +S B -intersection

[0064] The final calculated IOU result is:

[0065]

[0066] 2) If the registration result meets the accuracy requirements, the parking spaces are merged and output.

[0067] Set the medium-precision registration error threshold M. When implementing it, the length and width of the parking space perceived at the current moment are mainly used. Set the IOU range obtained by setting the length and width error of the historically registered parking space within ±0.5m (decimeter-level accuracy). If the registration error is within the error range of the threshold M, the registration is considered successful, and the id and coordinates are merged (the default id is the historical tracking result id, and the coordinates are the currently perceived coordinates). Otherwise, the registration is considered to have failed, and the next step of low-precision parking space tracking is continued.

[0068] The steps of low-precision parking space tracking merging in step S3 are as follows:

[0069] 1) Calculate the parking space center point for the parking space corner points obtained in step S1;

[0070] Based on the coordinates of the four corner points of the parking space corresponding to id, the average value is taken as the center coordinate of the parking space corresponding to id PC = (x c ,y c ,z c ).

[0071] 2) Perform nearest neighbor calculation on the center points of all parking spaces;

[0072] Based on the transformation matrix Convert the historically tracked parking space center coordinate information to the current vehicle body coordinate system, and calculate the Euclidean distance with each parking space ID at the current moment. The formula is as follows:

[0073] Euclidean distance calculation formula: Where i is the ID of the parking space sensed at the current moment, j is the ID corresponding to the historical tracking result, PC i is the coordinate of the center of the i-th parking space, PC′ j Convert the center of the j-th tracking result parking space to the coordinates of the vehicle body coordinate system at the current moment;

[0074] Keep the minimum result of Euclidean distance calculation as the registration error and proceed to the next step.

[0075] 3) If the nearest neighbor result meets the accuracy requirements, the parking spaces are merged and output.

[0076] A low-precision registration error threshold L is set. When it is implemented, it is set to less than 1.1m, that is, meter-level registration accuracy. When the registration error is less than the threshold L, the registration is considered successful, and the id and coordinates are merged (the default id is the historical tracking result id, and the coordinates are the current moment perception coordinates). Otherwise, the registration is considered to have failed, and a new parking space is created for the current parking space, which is then output and saved to the historical tracking parking space, waiting for the next moment conversion and registration steps.

[0077] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A real-time adaptive surround-view parking space tracking method for autonomous driving parking, characterized in that: The following steps are involved: S1. Verify the visual surround perception results and update the conversion matrix based on vehicle driving information; S2. For the successfully verified visual perception parking spaces, historical tracking parking spaces and conversion matrices, the parking spaces are converted into point clouds and aligned to the current moment; S3, the parking spaces in the point cloud are tracked in three different precision ranges: high, medium and low. The successfully tracked parking spaces are merged to obtain the tracked parking spaces. Otherwise, the next precision range is entered for parking space tracking. For all tracking failures, new parking spaces are created. S4: Use the current parking space tracking result as the historical tracking parking space at the next moment, and repeat the above steps S1-S3, continuously obtain the latest parking space tracking results.

2. The method for real-time adaptive surround-view parking space tracking for autonomous driving parking according to claim 1, characterized in that: The steps to update the transformation matrix based on vehicle driving information are as follows: 1) Determine the validity of vehicle driving information, mainly including whether the input time and parameter range are consistent; 2) Calculate the yaw angle, coordinates and transformation matrix of the vehicle based on the starting position through the vehicle speed and yaw angular velocity in the vehicle driving information.

3. The method for real-time adaptive surround-view parking space tracking for autonomous driving parking according to claim 1, characterized in that: The steps for verifying the visual surround perception results are as follows: Determine the validity of the parking space in the perception results, mainly including whether the parking space coordinate range is in compliance.

4. The method for real-time adaptive surround-view parking space tracking for autonomous driving parking according to claim 1, characterized in that: The steps of point clouding and aligning the parking spaces to the current moment are as follows: 1) Track the current and historical parking space transformations based on the vehicle body transformation matrix.

5. The method for real-time adaptive surround-view parking space tracking for autonomous driving parking according to claim 1, characterized in that: Among the three parking space tracking with different accuracy ranges of high, medium and low, high-precision parking space tracking is an ICP tracking method based on parking space corner points, and its perception accuracy is less than 0.1m; Medium-precision parking space tracking is a tracking method based on IOU, and its perception accuracy is less than 0.5m. Low-precision parking space tracking is a tracking method based on the parking space center, and its perception accuracy is less than 1.1m.

6. The method for real-time adaptive surround-view parking space tracking for automatic driving parking according to claim 5, characterized in that: The steps of high-precision parking space tracking are as follows: 1) Set a high-precision registration error threshold H. For the same vacant parking space, the vehicle perceives the parking space visually at the current time t and the previous time t-1 respectively. The corner coordinates of the parking space at time t-1 are converted to the vehicle body system at time t through the conversion matrix. Then, the adjacent points are iteratively registered under the vehicle body system at time t. 2) If the registration error is less than the threshold H, the registration result is deemed to meet the accuracy requirements, and the di and coordinates of the parking space are merged and output, where the id is taken at time t-1 and the coordinates are taken at time t.

7. The method for real-time adaptive surround-view parking space tracking for autonomous driving parking according to claim 5, characterized in that: The steps of medium-precision parking space tracking are as follows: 1) Set the medium-precision registration error threshold M, establish a local coordinate system with the two vertical edges of the parking space to be matched at the current moment, obtain the local coordinate system conversion matrix, and convert all the corner points of the parking space to be matched to the local coordinate system based on the local coordinate system conversion matrix; 2) After converting all the corner points of the parking spaces to be matched into the local coordinate system, perform the intersection-over-union (IOU) registration calculation; 3) If the registration error is within the error range of the threshold M, the registration is considered successful, and the id and coordinates are merged, where the id is the historical tracking result id and the coordinates are the perceived coordinates at the current moment.

8. The method for real-time adaptive surround-view parking space tracking for automatic driving parking according to claim 5, characterized in that: The steps of low-precision parking space tracking are as follows: 1) Calculate the parking space center point for the parking space corner points obtained in step S1; 2) Perform nearest neighbor calculation on the center points of all parking spaces; 3) If the nearest neighbor result meets the accuracy requirement, the parking spaces are merged and output; otherwise, the parking spaces are newly created and output.

9. A computer program product, comprising a computer program, characterized in that When executed, the program is used to implement each step of the real-time adaptive surround-view parking space tracking method for automatic driving parking according to any one of claims 1 to 8.

10. A vehicle, comprising a parking space tracking device, characterized in that: The parking space tracking device stores a computer program, which, when executed, is used to implement each step of the automatic driving parking real-time adaptive surround view parking space tracking method according to any one of claims 1 to 8.