Parking space detection processing method and device, electronic equipment and storage medium

Abstract

The invention relates to a parking space detection processing method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a target ground marking point cloud under an odom coordinate system; the target ground marking point cloud has a corresponding parking space true value; obtaining the current conversion information of the detection frame at the corresponding time, and converting the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame under the vehicle body coordinate system; determining a matched parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame; and training based on a preset processing algorithm and the matched parking space value pair to obtain a parking space detection processing model, so that the real-time parking space detection result of the vehicle is adjusted based on the parking space detection processing model. By adopting the above technical scheme, the parking space truth values of a large number of detection frames can be quickly obtained, so that the parking space detection processing model is quickly and effectively trained to adjust the real-time parking space detection result of the vehicle, and the parking space detection processing precision is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of vehicle intelligent driving technology, and in particular to a parking space detection processing method, device, electronic device and storage medium. Background Art

[0002] Usually, vehicles have low-speed intelligent driving functions, such as automatic parking, and one of the key technologies is the detection of parking spaces. The detection accuracy of parking spaces has a great impact on subsequent planning and control. Large deviations may even cause collision risks, seriously affecting the user experience of the function.

[0003] At present, parking space detection is generally based on image stitching using four surround-view cameras to obtain a surround-view stitched image, thereby detecting parking spaces. Most current detection schemes are based on deep learning, which uses a large amount of parking space annotation data on the stitched image to train a better model, detect parking space information on the stitched image in real time, and obtain the accuracy of the detection model by comparing statistics with the true value on the stitched image.

[0004] However, the method of determining the true value of the parking space through RTK (Real-time kinematic) technology is rather cumbersome and cannot obtain a large amount of data. In addition, the parking space detection solution based on the mosaic image is heavily dependent on the quality of the mosaic image, which in turn depends on the calibration of the camera. The static calibration result between calibrations can only guarantee the stitching effect when the vehicle body has no posture change on a flat road surface at that time. Therefore, during the movement of the vehicle, the acceleration and deceleration, speed bumps, rapid turns, etc. will cause changes in the body posture, which will cause the mosaic image to be distorted, thereby directly affecting the parking space detection result based on the mosaic image. Summary of the invention

[0005] In order to solve the above technical problem or at least partially solve the above technical problem, at least one embodiment of the present disclosure provides a parking space detection processing method, device, electronic device and storage medium.

[0006] In a first aspect, the present disclosure provides a parking space detection processing method, comprising:

[0007] Obtaining a target ground marking point cloud in an odom (odometer) coordinate system; wherein the target ground marking point cloud has a corresponding parking space true value;

[0008] Acquire current conversion information at a time corresponding to a detection frame, and convert the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in a vehicle body coordinate system;

[0009] Acquire a parking space detection value of the detection frame, and determine a matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame;

[0010] Based on the preset processing algorithm and the matching parking space value pair, training is performed to obtain a parking space detection processing model, so that the real-time parking space detection result of the vehicle is adjusted based on the parking space detection processing model.

[0011] Optionally, before obtaining the target ground marking point cloud in the odom coordinate system, the method further includes:

[0012] Acquire multiple frames of laser point cloud data based on a laser radar, perform ground point segmentation on each frame of the laser point cloud data, and acquire multiple frames of ground point cloud data;

[0013] Extracting point clouds from each frame of the ground point cloud data based on the reflection intensity corresponding to each frame of the ground point cloud data to obtain multiple frames of ground marking point clouds;

[0014] The ground marking point cloud of each frame is converted by laser radar parameters to obtain ground marking point clouds in multiple vehicle body coordinate systems;

[0015] The ground marking point cloud in each vehicle body coordinate system is converted by using the conversion information stored in odom to obtain the ground marking point clouds in multiple odom coordinate systems;

[0016] After superimposing and downsampling the ground marking point clouds in the multiple odom coordinate systems, the target ground marking point cloud in the odom coordinate system is obtained, and the target ground marking point cloud in the odom coordinate system is three-dimensionally annotated to obtain the true value of the parking space corresponding to the target ground marking point cloud in the odom coordinate system.

[0017] Optionally, the obtaining of current conversion information at a time corresponding to the detection frame includes:

[0018] Obtain first conversion information of a first time and second conversion information of a second time stored in odom; wherein the detection frame corresponding time is greater than the first time and less than the second time;

[0019] Interpolation processing is performed on the first conversion information and the second conversion information to obtain current conversion information at a time corresponding to the detection frame.

[0020] Optionally, after obtaining the true value of the parking space of the detection frame in the vehicle body coordinate system, the method further includes:

[0021] Filtering the parking space true value of the detection frame based on the region of interest of the detection frame to obtain the target parking space true value of the detection frame;

[0022] The determining a matching parking space value based on the parking space true value of the detection frame and the parking space detection value of the detection frame includes:

[0023] The matching parking space value is determined based on the target parking space true value of the detection frame and the parking space detection value of the detection frame.

[0024] Optionally, the determining the matching parking space value pair based on the target parking space true value of the detection frame and the parking space detection value of the detection frame includes:

[0025] Calculate the degree of overlap between the target parking space true value of the detection frame and the parking space detection value based on a preset matching algorithm;

[0026] The target parking space true value and the parking space detection value corresponding to the overlap degree being greater than a preset threshold are used as the matching parking space value pair.

[0027] Optionally, the training based on the preset processing algorithm and the matching parking space value pair to obtain a parking space detection processing model includes:

[0028] The parking space detection value in the matching parking space value pair and the vehicle posture angle corresponding to the detection frame stored in odom are used as input data of the preset regression model, and the parking space position deviation value determined by the matching parking space value pair is used as output data of the preset regression model for training to obtain the parking space detection processing model.

[0029] Optionally, the method further includes:

[0030] Obtaining the current posture information of the vehicle and the real-time detection result of the parking space;

[0031] Inputting the posture information and the real-time parking space detection result into the parking space detection processing model to obtain a parking space position deviation value;

[0032] The real-time parking space detection result is adjusted based on the parking space position deviation value.

[0033] In a second aspect, the present disclosure provides a parking space detection processing device, comprising:

[0034] The first acquisition module is used to acquire a target ground marking point cloud in an odom coordinate system; wherein the target ground marking point cloud has a corresponding parking space true value;

[0035] The second acquisition module is used to acquire the current conversion information at the corresponding time of the detection frame;

[0036] A first conversion module, configured to convert the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in a vehicle body coordinate system;

[0037] A third acquisition module is used to obtain the parking space detection value of the detection frame;

[0038] A determination module, configured to determine a matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame;

[0039] The processing module is used to perform training based on a preset processing algorithm and the matching parking space value pair to obtain a parking space detection processing model, so as to adjust the real-time parking space detection result of the vehicle based on the parking space detection processing model.

[0040] In a third aspect, the present disclosure provides an electronic device, including: a processor and a memory;

[0041] The processor is used to execute any of the parking space detection processing methods provided in the embodiments of the present disclosure by calling the program or instruction stored in the memory.

[0042] In a fourth aspect, the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores a program or instruction, wherein the program or instruction enables a computer to execute any of the parking space detection processing methods provided in the embodiments of the present disclosure.

[0043] In a fifth aspect, the present disclosure provides a computer program product, which is used to execute any of the parking space detection and processing methods provided in the embodiments of the present disclosure.

[0044] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has at least the following advantages:

[0045] In the disclosed embodiment, a target ground marking point cloud in the odom coordinate system is obtained; the target ground marking point cloud has a corresponding parking space true value; the current conversion information at the corresponding time of the detection frame is obtained, and the parking space true value is converted based on the current conversion information to obtain the parking space true value of the detection frame in the vehicle body coordinate system; the parking space detection value of the detection frame is obtained, and a matching parking space value pair is determined based on the parking space true value of the detection frame and the parking space detection value of the detection frame; training is performed based on a preset processing algorithm and matching parking space value pairs to obtain a parking space detection processing model, so that the real-time parking space detection result of the vehicle is adjusted based on the parking space detection processing model. By adopting the above technical solution, the parking space true value of a large number of detection frames can be quickly obtained based on the point cloud method, reducing the annotation workload; the parking space detection processing model can predict the parking space detection deviation caused by the change of the vehicle body posture, and compensate the parking space detection result in real time, ensuring real-time performance while improving the accuracy and robustness of the parking space detection processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0048] Figure 1 A flowchart of a parking space detection processing method provided by an exemplary embodiment of the present disclosure;

[0049] Figure 2 A flowchart of a parking space detection processing method provided by another exemplary embodiment of the present disclosure;

[0050] Figure 3 An example diagram of a method for obtaining a parking space true value provided in a specific embodiment of the present disclosure;

[0051] Figure 4 An example diagram of a parking space detection and adjustment method provided in a specific embodiment of the present disclosure;

[0052] Figure 5 A schematic diagram of the structure of a parking space detection and processing device provided in one embodiment of the present disclosure. DETAILED DESCRIPTION

[0053] In order to more clearly understand the above-mentioned purposes, features and advantages of the present disclosure, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understandable that the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments, and the specific embodiments described herein are only used to explain the present disclosure, rather than to limit the present disclosure. In the absence of conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in the field belong to the scope of protection of the present disclosure.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0055] At present, the detection of parking spaces is generally based on image stitching of four-way surround view cameras to obtain surround view stitching images, thereby detecting parking spaces; most of the current detection schemes are based on deep learning, using a large amount of parking space annotation data on the stitching map to train a better model, detect the parking space information on the stitching map in real time, and compare and count with the true value on the stitching map to obtain the accuracy of the detection model, but it is difficult to obtain a quantitative indicator for the actual perceived parking space position and direction. In addition, the way to obtain the true value of the parking space is usually to directly use RTK technology to mark the parking space to obtain the location of the parking space, but this method is cumbersome and cannot obtain a large amount of data, and the parking space detection scheme based on the stitching map is heavily dependent on the quality of the stitching map, and the quality of the stitching map depends on the calibration of the camera. The static calibration result between calibrations can only guarantee the stitching effect when the body has no posture change on the flat road at that time; therefore, during the movement of the vehicle, due to acceleration and deceleration, speed bumps, rapid steering, etc., the body posture changes will cause the stitching image to be distorted, which will directly affect the parking space detection result based on the stitching image.

[0056] It is also understandable that the camera dynamic calibration technology can be used to obtain the body posture and ground features and other real-time camera external parameters to correct the stitching effect in real time. However, this method requires obtaining feature point information and mapping the position of each pixel of the four fisheye images to the stitching image, which requires a lot of computing resources and is difficult to meet real-time requirements. In addition, when there are insufficient feature points or large errors, the calibration accuracy is also difficult to meet the requirements.

[0057] In view of the above problems, the present invention provides a parking space detection processing method, which improves the accuracy and robustness of parking space detection by using a parking space line marking method based on a laser radar and a method of compensating for parking space detection deviations caused by distortion of a spliced ​​image during movement in real time. Specifically, a ground point cloud segmentation, reflection intensity screening and multi-frame point cloud superposition are used to obtain a ground marking point cloud, and parking space marking is performed based on the point cloud to obtain the parking space true value in the positioned odom coordinate system, and the parking space true value is projected to the vehicle body coordinate system corresponding to each detection frame time, thereby obtaining the parking space true value of a large number of detection frames, reducing the marking workload. , and based on the acquired parking space true value and combined with a large amount of actual parking space detection result data and vehicle body posture data, the parking space detection processing model is trained to obtain the deviation mapping relationship between different positions of the detection frame and the actual position of the vehicle body coordinates when the vehicle body postures are different, so that the parking space detection results can be compensated in real time, and the parking space true value in the vehicle body coordinate system can be efficiently obtained, including the parking space position, orientation, type and occupancy attributes for quantitative evaluation, and the parking space detection deviation caused by the distortion of the splicing image caused by the change of the vehicle body posture can be improved in real time without significantly increasing the computing cost, thereby improving the accuracy and robustness of parking space detection.

[0058] The specific implementation of the parking space detection processing method, device, electronic device and storage medium disclosed in the present invention is explained in detail below with reference to the accompanying drawings.

[0059] Figure 1 This is a flow chart of a parking space detection processing method provided in an exemplary embodiment of the present disclosure. The method may be executed by a parking space detection processing device provided in an embodiment of the present disclosure. The parking space detection processing device may be implemented in software and / or hardware.

[0060] like Figure 1 As shown, the parking space detection processing method may include the following steps:

[0061] Step 101, obtaining a target ground marking point cloud in an odom coordinate system; wherein the target ground marking point cloud has a corresponding parking space true value.

[0062] Among them, the odom coordinate system refers to the vehicle's relative motion coordinate system. For example, if the vehicle starts to move, the coordinate system is established by setting the vehicle's position as the origin.

[0063] In the disclosed embodiment, the target ground marking point cloud refers to a point cloud set of ground markings. It can be understood that multiple frames of laser point cloud data can be acquired based on the laser radar on the vehicle, and the ground marking point cloud can be extracted from each frame of laser point cloud data. All ground marking point clouds are converted by laser radar parameters to obtain the ground marking point cloud in the vehicle body coordinate system, and the ground marking point cloud in the vehicle body coordinate system is converted by the conversion information stored in odom to obtain the ground marking point cloud in the odom coordinate system. All ground marking point clouds in the odom coordinate system are processed to obtain the target ground marking point cloud in the odom coordinate system, and the target ground marking point cloud in the odom coordinate system is labeled, so that the target ground marking point cloud has the corresponding parking space true value.

[0064] Among them, the true value of the parking space may include the parking space position coordinate value (such as the two angular coordinate values ​​of the parking space entrance), orientation (the angle relative to the vehicle, generally the counterclockwise direction of the X-axis is positive), type (referring to an attribute of the parking space itself, such as parallel parking space, vertical parking space, inclined parking space) and occupancy attribute, etc.

[0065] In the disclosed embodiment, when receiving a parking space detection processing instruction or a parking space real-time detection processing process, a target ground marking point cloud in the odom coordinate system can be obtained, and the target ground marking point cloud has a corresponding parking space true value.

[0066] Step 102, obtaining the current conversion information at the time corresponding to the detection frame, and converting the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in the vehicle body coordinate system.

[0067] In the disclosed embodiment, the detection frame refers to the surround stitching image obtained by stitching images based on the camera (such as a four-way surround camera), and the detection frames corresponding to different times are different, that is, each detection frame has a corresponding detection time, which can be understood as the detection frame corresponding time. It can also be understood that during the parking detection process, the vehicle moves in real time, so the conversion information (vehicle position and attitude angle) of the vehicle at different detection frame corresponding times is different, and the current conversion information refers to the current vehicle position and current attitude angle of the vehicle at the detection frame corresponding time.

[0068] It can be understood that odom stores the conversion information corresponding to each time of the vehicle according to a preset period, and the current conversion information at the time corresponding to the detection frame can be directly obtained from the conversion information stored in odom. The current conversion information at the time corresponding to the detection frame can also be obtained by interpolating the first conversion information of the first time and the second conversion information of the second time stored in odom, wherein the time corresponding to the detection frame is greater than the first time and less than the second time.

[0069] In the disclosed embodiment, the current conversion information at the corresponding time of the detection frame is obtained to perform coordinate system conversion on the parking space true value, so as to obtain the parking space true value of the detection frame in the vehicle body coordinate system. The parking space true value includes the parking space position coordinate value (such as the two corner coordinate values ​​of the parking space entrance), direction (the angle of the parking space relative to the vehicle, generally the counterclockwise direction of the X-axis is positive), type (referring to an attribute of the parking space itself, such as parallel parking space, vertical parking space, inclined parking space) and occupancy attribute, etc., which are used for quantitative evaluation.

[0070] Step 103: Acquire the parking space detection value of the detection frame, and determine a matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame.

[0071] In the embodiments of the present disclosure, there are many ways to obtain the parking space detection value of the detection frame, such as predicting / detecting the detection frame based on a deep learning model or algorithm to obtain the parking space detection value, that is, the parking space detection value can be understood as the parking space position coordinate value predicted / detected based on the detection frame.

[0072] In the embodiments of the present disclosure, there are many ways to determine a matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame. The degree of overlap between the parking space true value of the detection frame and the parking space true value of the detection frame can be directly calculated, and the parking space true value and the parking space detection value corresponding to the degree of overlap greater than a preset threshold are used as matching parking space value pairs; the parking space true value of the detection frame can also be screened based on the area of ​​interest of the detection frame to obtain the target parking space true value of the detection frame, and the degree of overlap between the target parking space true value and the parking space detection value of the detection frame is calculated, and the target parking space true value and the parking space detection value corresponding to the degree of overlap greater than the preset threshold are used as matching parking space value pairs.

[0073] Step 104 , training is performed based on a preset processing algorithm and matching parking space value pairs to obtain a parking space detection processing model, so as to adjust the real-time parking space detection result of the vehicle based on the parking space detection processing model.

[0074] In the disclosed embodiment, the processing algorithm can be selected and set according to the actual application needs, such as a regression analysis algorithm, etc. There are many ways to obtain a parking space detection processing model based on the preset processing algorithm and the matching parking space value pair for training. As an example, the parking space detection value in the matching parking space value pair and the vehicle posture angle corresponding to the detection frame stored in odom are used as input data of the preset regression model, and the parking space position deviation value determined by the matching parking space value pair is used as the preset regression model output data for training to obtain the parking space detection processing model; as another example, the parking space detection value in the matching parking space value pair and the vehicle posture angle corresponding to the detection frame stored in odom are input into the neural network to be trained for training to obtain a predicted deviation value, and the network parameters of the neural network to be trained are continuously adjusted based on the error value between the predicted deviation value and the parking space position deviation value determined by the matching parking space value pair until the error value meets a certain value to obtain the parking space detection processing model.

[0075] Furthermore, the current posture information of the vehicle and the real-time detection result of the parking space are obtained and input into the parking space detection processing model to obtain the parking space position deviation value, and the real-time detection result of the parking space is adjusted based on the parking space position deviation value.

[0076] The parking space detection processing method of the disclosed embodiment obtains the target ground marking point cloud in the odom coordinate system; the target ground marking point cloud has the corresponding parking space true value; obtains the current conversion information at the corresponding time of the detection frame, converts the parking space true value based on the current conversion information, and obtains the parking space true value of the detection frame in the vehicle body coordinate system; obtains the parking space detection value of the detection frame, and determines the matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame; trains based on the preset processing algorithm and the matching parking space value pair to obtain the parking space detection processing model, so as to adjust the real-time parking space detection result of the vehicle based on the parking space detection processing model. By adopting the above technical solution, the parking space true value of a large number of detection frames can be quickly obtained based on the point cloud method, reducing the annotation workload; the parking space detection processing model can predict the parking space detection deviation caused by the change of the vehicle body posture, and compensate the parking space detection result in real time, ensuring the real-time performance while improving the accuracy and robustness of the parking space detection processing.

[0077] Figure 2 A flowchart of a parking space detection processing method provided by another exemplary embodiment of the present disclosure is provided. The method may be executed by a parking space detection processing device provided by an embodiment of the present disclosure. The parking space detection processing device may be implemented by software and / or hardware.

[0078] like Figure 2As shown, the parking space detection processing method may include the following steps:

[0079] Step 201, based on the laser radar, multiple frames of laser point cloud data are acquired, ground point segmentation is performed on each frame of laser point cloud data, multiple frames of ground point cloud data are acquired, point cloud extraction is performed on each frame of ground point cloud data based on the reflection intensity corresponding to each frame of ground point cloud data, and multiple frames of ground marking point clouds are obtained.

[0080] In the disclosed embodiment, at least one laser radar is installed on the vehicle, multiple frames of laser point cloud data are acquired during parking space detection, and ground point segmentation is performed on each frame of laser point cloud data through a point cloud segmentation algorithm to acquire each frame of ground point cloud data.

[0081] In the parking space detection scenario of the embodiment of the present disclosure, the parking space has a corresponding marking line. It can also be understood that the reflection intensities of different point cloud data are different. Therefore, the reflection intensity of the marking point cloud is different from the reflection intensity of other point clouds on the ground. Therefore, point cloud extraction can be performed according to the reflection intensity corresponding to each point cloud data in each frame of ground point cloud data, such as taking the point cloud with a reflection intensity greater than or equal to a preset intensity threshold as each frame of ground marking point cloud.

[0082] Step 202, convert each frame of ground marking point cloud by laser radar parameters to obtain ground marking point clouds in multiple vehicle coordinate systems, and convert the ground marking point cloud in each vehicle coordinate system by conversion information stored in odom to obtain ground marking point clouds in multiple odom coordinate systems.

[0083] In the disclosed embodiment, each frame of ground marking point cloud obtained is point cloud data in the laser radar coordinate system. Therefore, it is necessary to convert each frame of ground marking point cloud by laser radar parameters to obtain the ground marking point cloud in the vehicle body coordinate system; wherein, the laser radar parameters are laser radar external parameters used to convert the point cloud in the laser radar coordinate system to the point cloud in the vehicle body coordinate system.

[0084] In the disclosed embodiment, odom is an odometer, also known as a driving mileage meter, which is an instrument used to measure the mileage of a vehicle. The odometer is usually installed on the dashboard of the vehicle. The mileage of the vehicle is calculated by recording the number of times the vehicle tires rotate or the distance the wheels rotate. The odometer function is usually implemented by a vehicle speed sensor built into the vehicle's transmission system, which can record information such as the real-time position and attitude angle of the vehicle.

[0085] Furthermore, the ground marking point cloud in each vehicle body coordinate system is transformed through the conversion information stored in odom to obtain the ground marking point cloud in multiple odom coordinate systems; wherein the conversion information includes the vehicle position and attitude angle as the conversion matrix to convert the ground marking point cloud in the vehicle body coordinate system to the ground marking point cloud in the odom coordinate system.

[0086] Step 203, after superimposing and downsampling the ground marking point clouds in multiple odom coordinate systems, a target ground marking point cloud in the odom coordinate system is obtained, and the target ground marking point cloud in the odom coordinate system is three-dimensionally annotated to obtain the true value of the parking space corresponding to the target ground marking point cloud in the odom coordinate system.

[0087] In the embodiments of the present disclosure, the frequency of point cloud data is usually relatively high, which leads to an increase in data volume and reduces data processing efficiency. Therefore, ground marking point clouds in multiple ODOM coordinate systems can be superimposed and downsampled to obtain a point cloud map composed of dense ground marking point clouds on a certain driving path, thereby reducing the frequency of point cloud data, reducing data volume, and improving processing efficiency.

[0088] Furthermore, the target ground marking point cloud in the odom coordinate system is three-dimensionally annotated to obtain the true value of the parking space corresponding to the target ground marking point cloud in the odom coordinate system; the true value of the parking space includes the parking space position coordinate value (such as the two corner coordinate values ​​of the parking space entrance), orientation (the angle of the parking space relative to the vehicle, generally the counterclockwise direction of the X-axis is positive), type (referring to an attribute of the parking space itself, such as parallel parking space, vertical parking space, inclined parking space) and occupancy attribute, etc., which are used for quantitative evaluation.

[0089] Step 204, obtain the first conversion information of the first time and the second conversion information of the second time stored in odom; wherein, the detection frame corresponding time is greater than the first time and less than the second time, interpolate the first conversion information and the second conversion information to obtain the current conversion information at the detection frame corresponding time.

[0090] Step 205, converting the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in the vehicle body coordinate system, filtering the parking space true value of the detection frame based on the region of interest of the detection frame to obtain the target parking space true value of the detection frame.

[0091] In the disclosed embodiment, odom stores conversion information corresponding to each time during the vehicle's driving process (including information such as the vehicle's real-time position and attitude angle), so the current conversion information at the corresponding time of the detection frame can be obtained by interpolation.

[0092] Specifically, the first conversion information of the first time and the second conversion information of the second time stored in odom are obtained, and the first conversion information and the second conversion information are interpolated to obtain the current conversion information at the time corresponding to the detection frame, and the time corresponding to the detection frame is greater than the first time and less than the second time, so as to convert the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in the vehicle body coordinate system.

[0093] For example, it is necessary to obtain the ground truth of the parking space (in the vehicle body coordinate system) for the stitched image of the current detection frame (assuming the time corresponding to the detection frame is t'), and the ground truth of the parking space in the odom coordinate system (with the coordinate origin being the odom origin) has been obtained previously. Therefore, based on the positions (relative to the odom origin) and attitude angles of the vehicle at the first time t0 and the second time t1 cached in the odometer odom (t0 < t' < t1), the position and attitude angle at t' can be obtained by interpolation. According to the position of the vehicle itself at t' (relative to the odom coordinate origin) and the ground truth of the parking space in the odom coordinate system, the ground truth of the parking space in the vehicle body coordinate system corresponding to the time t' of the detection frame can be obtained, that is, the ground truth of the parking space in the vehicle body coordinate system for the detection frame.

[0094] To improve the processing efficiency, the ground truth of the parking space in the detection frame can be screened according to the region of interest of the detection frame to obtain the target ground truth of the parking space in the detection frame, that is, the ground truth of the parking space within the region of interest of the detection frame will be processed subsequently.

[0095] Specifically, the ground truth of the parking space in the vehicle body coordinate system obtained for the detection frame can be understood as the ground truth of the parking space marked in the entire point cloud map area. However, since the distance of the detection frame (stitched image) is generally within a certain distance around the vehicle itself, such as 10m, it is necessary to screen according to the region of the detection frame. Among them, the region of the detection frame is the ROI (Region of Interest) region to further improve the processing efficiency and effect.

[0096] Step 206: Obtain the parking space detection value of the detection frame, calculate the overlap degree between the target ground truth of the parking space in the detection frame and the parking space detection value based on a preset matching algorithm, and use the target ground truth and the parking space detection value corresponding to the overlap degree greater than the preset threshold as the matching parking space value pair.

[0097] In the embodiments of the present disclosure, the parking space detection value of the detection frame can be parking space information predicted / detected based on deep learning, such as parking space position coordinate values (such as the coordinate values of two corners of the parking space entrance), orientation (the angle of the parking space relative to the vehicle itself, generally positive in the counterclockwise direction of the X axis), type (referring to an attribute of the parking space itself, such as a parallel parking space, a perpendicular parking space, an inclined parking space), and occupancy attribute, etc.

[0098] In the disclosed embodiment, the matching algorithm can be selected and set as needed, such as the Hungarian matching algorithm, which performs Hungarian matching on the parking space true value (i.e., the target parking space true value of the detection frame) and the parking space detection value (i.e., the detection result of the detection frame) within the ROI, and selects the target parking space true value and the parking space detection value corresponding to the overlap, i.e., IoU (Intersection over Union) greater than a certain preset threshold as the matching parking space value pair, and saves the position and attribute information (such as whether the parking space is occupied, the type of parking space, etc.) of the matched target parking space true value and the parking space detection value.

[0099] Step 207, using the parking space detection value in the matching parking space value pair and the vehicle posture angle corresponding to the detection frame stored in odom as input data of the preset regression model, and the parking space position deviation value determined by the matching parking space value pair as output data of the preset regression model for training, to obtain a parking space detection processing model.

[0100] In the disclosed embodiment, the preset regression model can be set as needed, such as a support vector machine model, and each pair of parking space position coordinate values, such as the coordinate values ​​of the two entrance corner points, and the vehicle attitude angle corresponding to each detection frame obtained from the odometer are obtained from the multiple pairs of parking space true values ​​and parking space detection values ​​that are successfully matched, including pitch (pitch angle) and roll (roll angle). For example, the vehicle attitude angle pitch, roll, and the position coordinates of the two entrance corner points of the parking space detection are used as input data, and the deviation value of the parking space true value and the parking space detection value is used as output data, and the input and output mapping relationship is regressed using methods such as support vector machines to obtain a parking space detection processing model.

[0101] Step 208, obtaining the current posture information and real-time detection results of the vehicle, inputting the posture information and the real-time detection results of the parking space into the parking space detection processing model, obtaining the parking space position deviation value, and adjusting the real-time detection results of the parking space based on the parking space position deviation value.

[0102] In the disclosed embodiment, the posture information and real-time detection results of the vehicle at the current moment can be obtained in real time, and the posture information and the real-time detection results of the parking space can be input into the parking space detection processing model to obtain the parking space position deviation value, and the real-time detection results of the parking space can be adjusted based on the parking space position deviation value. For example, the parking space position deviation value is the coordinate deviation value of the parking space entrance corner point 1 X1 and Y1, and the real-time detection results such as the coordinates X0 and Y0 of the parking space entrance corner point 1 are added or subtracted from the coordinate deviation values ​​X1 and Y1 respectively to obtain the coordinate value of the actual parking space entrance corner point 1, so as to determine the parking space position according to the adjusted coordinate value, that is, the parking space detection processing model can be imported into the parking space perception post-processing module, and the real-time detection results (that is, the parking space corner point coordinate values) given by the parking space detection processing model can be compensated for the deviation according to the posture of the vehicle at different times, so as to obtain the parking space entrance corner point compensation information detected under the current vehicle position and posture, which can improve the accuracy of parking space detection, meet vehicle use requirements, and enhance user experience.

[0103] Based on the above description, the parking space detection processing method of the embodiment of the present disclosure preferably obtains the true value of the parking space of the detection frame based on the point cloud, and uses the true value of the parking space based on the detection frame for training the model to perform real-time compensation for the parking space detection result.

[0104] As an example, Figure 3 The method for obtaining the true value of parking spaces shown in the figure includes step 3.1 inputting single-frame laser point cloud data; step 3.2 ground segmentation of point cloud; step 3.3 extracting ground marking point cloud; step 3.4 converting the ground marking point cloud to the vehicle body coordinate system through the laser radar external parameter; step 3.5 converting the ground marking point cloud to the odom coordinate system through the vehicle position and posture; step 3.6 downsampling the ground marking point cloud in the odom coordinate system; step 3.7 fusing all ground marking point clouds; step 3.8 marking the parking spaces on the fused point cloud map.

[0105] Specifically, based on the acquired single-frame laser point cloud data, ground point segmentation is performed on each frame of laser point cloud data, and the ground sign and marking point cloud is extracted by reflection intensity (different objects have different reflection intensities); the extracted sparse point cloud is transferred to the vehicle body coordinate system through the laser radar external parameter, and then the point cloud is transferred to the Odom coordinate system through the vehicle position and attitude angle provided by Odom, and then superimposed with the point cloud processed in the previous frame and down-sampled, and finally a point cloud map composed of dense ground marking point clouds on a certain driving path is obtained, and finally three-dimensional annotation is performed based on the above point cloud to obtain the true value of the parking space in the Odom coordinate system, that is, the location of the parking space and other information.

[0106] As another example, Figure 4The parking space detection deviation real-time compensation method shown in the figure includes step 4.1 interpolating the cached odom positioning data to obtain the vehicle position and attitude angle at the time corresponding to the detection result of the current detection frame; step 4.2 converting the parking space true value obtained in the odom coordinate system to the vehicle body coordinate system at the time of the current detection frame; step 4.3 filtering the parking space true value after the above conversion according to the ROI of the current detection frame, and the parking space true value within the ROI is the parking space true value of the current detection frame, which can be used for parking space perception evaluation and subsequent association; step 4.4 obtaining the current detection frame (surround view stitching image), and performing parking space detection on the current detection frame to obtain the parking space detection result; step 4.5 Matching the true value of the parking space with the predicted parking space, that is, matching the true value of the parking space within the ROI with the parking space detection result (such as Hungarian matching), screening out the optimal matching pair with IoU (intersection over union) greater than a certain threshold, and saving the position and attribute information of the matched true value of the parking space and the parking space detection result; Step 4.6 saves the multi-frame matching results until the end, that is, continues the detection of the next frame, repeats the above steps to save the multi-frame detection results and the corresponding parking space true value; Step 4.7 trains the model, extracts the vehicle's posture angle pitch, roll, and the two corner point positions x and y of the entrance of the parking space detection from the above data as input data, and uses the deviation delta x and delta y between the true value of the parking space entrance position and the detection result as output data, and uses methods such as support vector machines to regress the mapping relationship between the input data and the output data, and saves the model; Step 4.8 real-time compensation of corner points, imports the model into the parking space perception post-processing module, and according to the posture of the vehicle at different times, performs deviation compensation on the parking space corner point coordinate values ​​given by the model, and obtains the parking space entrance corner point compensation information detected under the current vehicle position and posture.

[0107] The parking space detection and processing method of the disclosed embodiment obtains multiple frames of laser point cloud data based on a laser radar, performs ground point segmentation on each frame of laser point cloud data, obtains multiple frames of ground point cloud data, extracts point clouds from each frame of ground point cloud data based on the reflection intensity corresponding to each frame of ground point cloud data, obtains multiple frames of ground marking point clouds, converts each frame of ground marking point clouds by laser radar parameters, obtains ground marking point clouds in multiple vehicle body coordinate systems, converts the ground marking point clouds in each vehicle body coordinate system by conversion information stored in odom, obtains ground marking point clouds in multiple odom coordinate systems, superimposes and downsamples the ground marking point clouds in the multiple odom coordinate systems, obtains the target ground marking point cloud in the odom coordinate system, performs three-dimensional annotation on the target ground marking point cloud in the odom coordinate system, obtains the true value of the parking space corresponding to the target ground marking point cloud in the odom coordinate system, and obtains the first conversion information of the first time and the second conversion information of the second time stored in odom; wherein the detection frame corresponding time is greater than the first time and less than the second time, the first conversion information is converted to the first conversion information. The present invention relates to a method for detecting a parking space in a vehicle body coordinate system, wherein the present invention relates to a parking space detection process ... In this way, the true value of the parking space can be obtained quickly, and a small amount of parking space annotation results can be used to generate the true value of the parking space corresponding to a large number of data frames, which can be used for parking space perception evaluation and data analysis. Without changing the existing model and splicing diagram, the detection deviation caused by the change of the vehicle body posture is predicted, and the parking space detection result is compensated in real time. Compared with real-time dynamic calibration and regeneration of the splicing mapping table, there is no excessive increase in calculation amount, and the real-time performance can be guaranteed and the accuracy meets the requirements.

[0108] In order to implement the above embodiments, the present disclosure further provides a parking space detection processing device, which can be implemented by software and / or hardware.

[0109] Figure 5A schematic diagram of the structure of a parking space detection processing device provided in an embodiment of the present disclosure is shown in FIG. Figure 4 As shown, the parking space detection processing device 50 may include: a first acquisition module 510 , a second acquisition module 520 , a first conversion module 530 , a third acquisition module 540 , a determination module 550 and a first sending module processing module 560 .

[0110] The first acquisition module 510 is used to acquire a target ground marking point cloud in an odom coordinate system; wherein the target ground marking point cloud has a corresponding parking space true value;

[0111] The second acquisition module 520 is used to acquire the current conversion information at the time corresponding to the detection frame;

[0112] A first conversion module 530, configured to convert the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in a vehicle body coordinate system;

[0113] A third acquisition module 540 is used to acquire the parking space detection value of the detection frame;

[0114] A determination module 550, configured to determine a matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame;

[0115] The processing module 560 is used to perform training based on a preset processing algorithm and the matching parking space value pair to obtain a parking space detection processing model, so as to adjust the real-time parking space detection result of the vehicle based on the parking space detection processing model.

[0116] Optionally, the device further comprises:

[0117] An acquisition and segmentation module is used to acquire multiple frames of laser point cloud data based on a laser radar, and to perform ground point segmentation on each frame of the laser point cloud data to acquire multiple frames of ground point cloud data;

[0118] An extraction module, used to extract point cloud from each frame of the ground point cloud data based on the reflection intensity corresponding to each frame of the ground point cloud data, to obtain multiple frames of ground marking point clouds;

[0119] A second conversion module is used to first convert each frame of the ground marking point cloud by using laser radar parameters to obtain ground marking point clouds in multiple vehicle body coordinate systems;

[0120] A third conversion module is used to convert the ground marking point cloud in each vehicle body coordinate system through the conversion information stored in odom to obtain the ground marking point cloud in multiple odom coordinate systems;

[0121] A processing and annotation module is used to superimpose and downsample the ground marking point clouds in the multiple odom coordinate systems to obtain the target ground marking point cloud in the odom coordinate system, and to perform three-dimensional annotation on the target ground marking point cloud in the odom coordinate system to obtain the true value of the parking space corresponding to the target ground marking point cloud in the odom coordinate system.

[0122] Optionally, the second acquisition module 520 is specifically configured to:

[0123] Obtain first conversion information of a first time and second conversion information of a second time stored in odom; wherein the detection frame corresponding time is greater than the first time and less than the second time;

[0124] Interpolation processing is performed on the first conversion information and the second conversion information to obtain current conversion information at a time corresponding to the detection frame.

[0125] Optionally, the device further comprises:

[0126] A screening module, configured to screen the parking space true value of the detection frame based on the region of interest of the detection frame to obtain the target parking space true value of the detection frame;

[0127] The determination module 550 is specifically configured to:

[0128] The matching parking space value is determined based on the target parking space true value of the detection frame and the parking space detection value of the detection frame.

[0129] Optionally, the determining module 550 is further configured to:

[0130] Calculate the degree of overlap between the target parking space true value of the detection frame and the parking space detection value based on a preset matching algorithm;

[0131] The target parking space true value and the parking space detection value corresponding to the overlap degree being greater than a preset threshold are used as the matching parking space value pair.

[0132] Optionally, the processing module is specifically used to:

[0133] The parking space detection value in the matching parking space value pair and the vehicle posture angle corresponding to the detection frame stored in odom are used as input data of the preset regression model, and the parking space position deviation value determined by the matching parking space value pair is used as output data of the preset regression model for training to obtain the parking space detection processing model, so as to adjust the real-time parking space detection result of the vehicle based on the parking space detection processing model.

[0134] Optionally, the device further comprises:

[0135] The first acquisition module is used to obtain the current posture information of the vehicle and the real-time detection result of the parking space;

[0136] An input module, used for inputting the posture information and the real-time parking space detection result into the parking space detection processing model to obtain a parking space position deviation value;

[0137] An adjustment module is used to adjust the real-time parking space detection result based on the parking space position deviation value.

[0138] The parking space detection processing device provided in the embodiment of the present disclosure can execute the parking space detection processing method provided in the embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method. For the contents not described in detail in the embodiment of the device of the present disclosure, reference can be made to the description in any method embodiment of the present disclosure.

[0139] The disclosed embodiment further provides an electronic device, including a processor and a memory; the processor calls a program or instruction stored in the memory to execute the steps of each embodiment of the aforementioned parking space detection processing method, which will not be described again here to avoid repeated description.

[0140] The embodiments of the present disclosure also provide a computer-readable storage medium, which is non-transitory and stores programs or instructions. The programs or instructions enable a computer to execute the steps of the embodiments of the parking space detection processing method described above. To avoid repeated description, they will not be repeated here.

[0141] The embodiments of the present disclosure also provide a computer program product, which is used to execute the steps of the embodiments of the aforementioned parking space detection processing method.

[0142] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0143] The above description is only a specific embodiment of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A parking space detection processing method, characterized in that: include: Obtaining a target ground marking point cloud in an odom coordinate system; wherein the target ground marking point cloud has a corresponding parking space true value; Acquire current conversion information at a time corresponding to a detection frame, and convert the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in a vehicle body coordinate system; Acquire a parking space detection value of the detection frame, and determine a matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame; Based on the preset processing algorithm and the matching parking space value pair, training is performed to obtain a parking space detection processing model, so that the real-time parking space detection result of the vehicle is adjusted based on the parking space detection processing model.

2. The method according to claim 1, characterized in that Before obtaining the target ground marking point cloud in the odom coordinate system, the method further includes: Acquire multiple frames of laser point cloud data based on a laser radar, perform ground point segmentation on each frame of the laser point cloud data, and acquire multiple frames of ground point cloud data; Extracting point clouds from each frame of the ground point cloud data based on the reflection intensity corresponding to each frame of the ground point cloud data to obtain multiple frames of ground marking point clouds; The ground marking point cloud of each frame is converted by laser radar parameters to obtain ground marking point clouds in multiple vehicle body coordinate systems; The ground marking point cloud in each vehicle body coordinate system is converted by using the conversion information stored in odom to obtain the ground marking point clouds in multiple odom coordinate systems; After superimposing and downsampling the ground marking point clouds in the multiple odom coordinate systems, the target ground marking point cloud in the odom coordinate system is obtained, and the target ground marking point cloud in the odom coordinate system is three-dimensionally annotated to obtain the true value of the parking space corresponding to the target ground marking point cloud in the odom coordinate system.

3. The method according to claim 1, characterized in that: The obtaining of the current conversion information at the time corresponding to the detection frame includes: Obtaining first conversion information of a first time and second conversion information of a second time stored in odom; wherein the detection frame corresponding time is greater than the first time and less than the second time; Interpolation processing is performed on the first conversion information and the second conversion information to obtain current conversion information at a time corresponding to the detection frame.

4. The method according to claim 1, characterized in that: After obtaining the true value of the parking space of the detection frame in the vehicle body coordinate system, the method further includes: Filtering the parking space true value of the detection frame based on the region of interest of the detection frame to obtain the target parking space true value of the detection frame; The determining a matching parking space value based on the parking space true value of the detection frame and the parking space detection value of the detection frame includes: The matching parking space value is determined based on the target parking space true value of the detection frame and the parking space detection value of the detection frame.

5. The method according to claim 4, characterized in that The determining the matching parking space value pair based on the target parking space true value of the detection frame and the parking space detection value of the detection frame includes: Calculate the degree of overlap between the target parking space true value of the detection frame and the parking space detection value based on a preset matching algorithm; The target parking space true value and the parking space detection value corresponding to the overlap degree being greater than a preset threshold are used as the matching parking space value pair.

6. The method according to claim 1, characterized in that The training based on the preset processing algorithm and the matching parking space value pair to obtain a parking space detection processing model includes: The parking space detection value in the matching parking space value pair and the vehicle posture angle corresponding to the detection frame stored in odom are used as input data of the preset regression model, and the parking space position deviation value determined by the matching parking space value pair is used as output data of the preset regression model for training to obtain the parking space detection processing model.

7. The method according to claim 1, characterized in that The method further comprises: Obtaining the current posture information of the vehicle and the real-time detection result of the parking space; Inputting the posture information and the real-time parking space detection result into the parking space detection processing model to obtain a parking space position deviation value; The real-time parking space detection result is adjusted based on the parking space position deviation value.

8. A parking space detection processing device, characterized in that: The device comprises: The first acquisition module is used to acquire a target ground marking point cloud in an odom coordinate system; wherein the target ground marking point cloud has a corresponding parking space true value; The second acquisition module is used to acquire the current conversion information at the corresponding time of the detection frame; A first conversion module, configured to convert the parking space true value based on the current conversion information to obtain the parking space true value of the detection frame in a vehicle body coordinate system; A third acquisition module is used to obtain the parking space detection value of the detection frame; A determination module, configured to determine a matching parking space value pair based on the parking space true value of the detection frame and the parking space detection value of the detection frame; The processing module is used to perform training based on a preset processing algorithm and the matching parking space value pair to obtain a parking space detection processing model, so as to adjust the real-time parking space detection result of the vehicle based on the parking space detection processing model.

9. An electronic device, characterized in that: include: Processor and memory; The processor is used to execute the parking space detection processing method according to any one of claims 1 to 7 by calling the program or instruction stored in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or an instruction, and the program or the instruction enables a computer to execute the parking space detection processing method according to any one of claims 1 to 7.

Images