Vehicle positioning method and system and storage medium

Abstract

The invention provides a vehicle positioning method and system and a storage medium, a target detection model is pre-constructed, and the vehicle positioning method comprises the following steps: obtaining vehicle pose data and all-round image data in real time, carrying out primary processing on the all-round image data based on preset image parameters to obtain an aerial view, detecting the aerial view through a target detection model to obtain target mark data, tracking the target mark data in real time based on a current map setting state, and matching the target mark data at any two continuous moments based on a first preset matching algorithm to obtain a matched aerial view; and correcting the vehicle pose data based on a matching result to obtain a final vehicle positioning result. According to the method provided by the invention, the problem of poor vehicle positioning precision in a special scene in the prior art is solved, the vehicle pose is optimized, the calculated amount in the vehicle positioning process is reduced, and the accuracy and reliability of a vehicle positioning result are improved.

Description

Technical Field

[0001] This application belongs to the technical field of positioning and navigation, and specifically relates to a vehicle positioning method, system, and storage medium. Background Art

[0002] With the acceleration of the urbanization process, the problem of difficult parking has become increasingly prominent. To solve this problem, intelligent parking systems have emerged. In the field of automatic parking, map construction and vehicle real-time positioning technology are one of the key technologies for automatic valet parking. In terms of maps, in scenarios with high-precision maps, the safety of unmanned driving devices in real road scenarios can be guaranteed to a certain extent. However, there are still many remote areas, underground parking lots, indoor parking buildings and other scenarios where perfect high-precision maps have not been established.

[0003] In terms of vehicle positioning, the commonly used solutions usually include: positioning methods based on GPS (Global Positioning System), positioning methods based on LiDAR (Light Detection and Ranging), positioning methods based on UWB (Ultra Wide Band) base stations, and visual positioning methods based on cameras for VSLAM (Visual Simultaneous Localization and Mapping). The GPS systems, lidar, and UWB base stations involved in the above methods generally have high costs in terms of hardware deployment. In the prior art, maps are usually directly constructed using semantic point clouds, which has problems such as large point cloud matching calculation amounts, difficult implementation, and poor positioning accuracy in special scenarios. Summary of the Invention

[0004] To solve the above technical problems, this application proposes a vehicle positioning method, system, and storage medium, aiming to solve the problem of poor positioning accuracy in special scenarios in the prior art, and reduce the large calculation amount and difficult implementation of map construction.

[0005] Specifically, this application proposes a vehicle positioning method, including:

[0006] Real-time acquisition of vehicle pose data and panoramic image data.

[0007] Based on preset image parameters, the panoramic image data is preliminarily processed to obtain a bird's-eye view.

[0008] The bird's-eye view is detected through a target detection model to obtain target marker data.

[0009] Based on the current map setting status, perform real-time tracking on the target landmark data, and match the target landmark data at any two consecutive moments based on a first preset matching algorithm, so as to correct the vehicle pose data based on the matching result and obtain the final vehicle positioning result.

[0010] In the above technical solution, the bird's-eye view is detected by the target detection model, and the obtained target landmark data is subjected to real-time tracking and matching, thereby updating the vehicle pose data, improving the accuracy of vehicle positioning, and improving the accuracy and reliability of the vehicle positioning result. By matching the target landmark data at any two consecutive moments through the first preset matching algorithm, the problem of large computational complexity and difficult implementation caused by directly using semantic point clouds to construct a map in the prior art is solved, the computational complexity in the vehicle positioning process is reduced, and the system deployment cost is reduced. By performing real-time tracking on the target landmark data, the timeliness and accuracy of the vehicle positioning result are ensured.

[0011] As an implementation manner, the preliminary processing at least includes distortion correction processing and inverse perspective transformation processing; the preliminary processing of the omnidirectional image data includes:

[0012] Perform distortion correction processing on the omnidirectional image data based on the preset image parameters, perform inverse perspective transformation processing on the omnidirectional image data after distortion correction processing to obtain the final omnidirectional image data; splice the final omnidirectional image data to obtain the bird's-eye view.

[0013] By performing distortion correction processing on the omnidirectional image data, the image quality of the omnidirectional image data is improved, and the authenticity and accuracy of the omnidirectional image data are improved. By performing inverse perspective transformation processing on the omnidirectional image data after distortion correction processing, the deviation caused by the perspective is eliminated, and a clearer and more intuitive final omnidirectional image data is obtained. The bird's-eye view spliced after distortion correction processing and inverse perspective transformation processing improves the accuracy of subsequent target detection and positioning, and the obtained accurate bird's-eye view can provide reliable and accurate scene information for the follow-up, reducing the risk of false detection and missed detection.

[0014] Furthermore, the pre-constructed target detection model includes:

[0015] Construct the target detection model through a deep learning algorithm; obtain historical target landmark data to train the target detection model through the historical target landmark data; perform optimization processing on the trained target detection model based on a preset optimization algorithm to obtain the final target detection model.

[0016] The target detection model is constructed through a deep learning algorithm and trained with historical target marker data, enabling the target detection model to handle complex environments and diverse target marker data, thereby improving the accuracy of target detection. Training with historical target marker data allows the target detection model to learn target marker data at different times and in different environments, enhancing the detection ability for various types of target marker data and improving the robustness of target detection. After continuously updating and accumulating historical target marker data, the target detection model can adapt to various scene changes, improving the adaptability of the target detection model. The target detection model is optimized through the preset optimization algorithm, further improving the detection efficiency and detection accuracy of the target detection model.

[0017] Further, the target marker data at least includes the type, position, and edge point cloud data of the target marker; the target detection model at least includes a target recognition model and a semantic segmentation model; the process of using the target detection model to detect the bird's-eye view and obtain target marker data includes:

[0018] Using the target recognition model to perform target recognition on the bird's-eye view to obtain the type and position of the target marker.

[0019] Using the semantic segmentation model to perform semantic segmentation on the bird's-eye view to extract the edge point cloud data based on the obtained semantic segmentation map.

[0020] By performing target recognition through the target recognition model, the type and position of the target marker can be accurately identified. Through the extraction of the edge point cloud data, it is possible to combine the type and position of the target marker, providing target marker data with higher precision for vehicle positioning, improving the detection accuracy of the edge point cloud data of the target detection model in complex scenarios, and avoiding misrecognition and missed recognition problems.

[0021] Further, the vehicle pose data at least includes the first vehicle pose and the mileage timestamp, and the surround-view image data at least includes the surround-view image and the image timestamp; before performing real-time tracking on the target marker data, it further includes:

[0022] Converting the target marker data to a vehicle-centered world coordinate system; in the world coordinate system, obtaining the mileage timestamp within a preset time threshold range based on the image timestamp; based on the time difference between the image timestamp and the mileage timestamp, performing a linear interpolation operation on the first vehicle pose corresponding to the mileage timestamp to obtain the second vehicle pose corresponding to the image timestamp.

[0023] By converting the target landmark data to the vehicle-centered world coordinate system and then obtaining the mileage timestamps within a preset time threshold range based on the image timestamp, the accuracy of the obtained mileage timestamps is ensured. Through linear interpolation of the first vehicle pose corresponding to the mileage timestamp based on the time difference between the image timestamp and the mileage timestamp, the time deviation of different timestamps is eliminated, ensuring the accuracy of the first vehicle pose.

[0024] Further, before correcting the vehicle pose data, it includes: obtaining the current map setting state. If the current map setting state is that no map is set, then the obtaining of the final vehicle positioning result includes:

[0025] Performing real-time tracking on the target landmark data at consecutive moments.

[0026] Performing preliminary matching on the target landmark data at any two consecutive moments, and based on the preliminary matching result, using a first preset matching algorithm to match the edge point cloud data to obtain the translation vector of the target landmark data at consecutive moments.

[0027] Constructing a vehicle pose constraint relationship at consecutive moments based on the translation vector, to correct the second vehicle pose based on the vehicle pose constraint relationship, and obtaining the final vehicle positioning result according to the correction result.

[0028] By performing real-time tracking on the target landmark data, the vehicle can update the position of the target landmark in real time, ensuring that the vehicle position is synchronized with the actual environment, thereby improving the accuracy of vehicle positioning. By performing preliminary matching on the target landmark data at any two consecutive moments, it enables quick judgment of the position change of the target landmark, provides preliminary position information for the subsequent matching algorithm, improves the efficiency of the matching algorithm, and improves the stability of the first preset matching algorithm. By using the first preset matching algorithm to match the edge point cloud data, the change of the target landmark in space can be obtained more accurately. By matching the edge point cloud, the detailed changes of the target landmark can be captured, thereby improving the accuracy of vehicle positioning. By establishing a vehicle pose constraint relationship between consecutive moments, the cumulative error and deviation can be effectively reduced. In the case of no preset map, this real-time vehicle pose constraint relationship can effectively improve the vehicle positioning accuracy and reduce the errors caused by long-term operation or high-speed driving.

[0029] Further, if the current map setting state is that a map is set, the obtaining of the final vehicle positioning result includes:

[0030] Converting the target landmark data to the map coordinate system to obtain the first position coordinate; using a second preset matching algorithm to match the target landmark data with the map to obtain the second position coordinate.

[0031] Obtain a rotation and translation vector based on the first position coordinate and the second position coordinate, so as to correct the pose of the second vehicle based on the rotation and translation vector, and obtain a final vehicle positioning result according to the correction result.

[0032] By converting the target landmark data into the map coordinate system, the first position coordinate is obtained. After conversion into the map coordinate system, the target landmark data is matched with the map through the second preset matching algorithm, which improves the alignment accuracy between the target landmark data and the map, thereby further reducing the positioning error. By matching the target landmark data with the set map through the second preset matching algorithm, the prior knowledge of the map can be effectively utilized, the computational burden of real-time positioning can be reduced, the matching efficiency and accuracy can be improved, and the accuracy of the final vehicle positioning is ensured. By correcting the pose of the second vehicle through the rotation and translation vector, it can be ensured that the final positioning result of the vehicle is highly consistent with the actual position, thereby improving the accuracy and robustness of vehicle positioning.

[0033] Further, after obtaining the final vehicle positioning result, it further includes:

[0034] Predict the vehicle running trajectory based on the final vehicle positioning result; perform corresponding scenario processing based on the vehicle running trajectory; wherein, the scenario processing at least includes smoothing processing and trajectory optimization processing.

[0035] Through trajectory prediction based on the final vehicle positioning result, the future movement path of the vehicle can be accurately predicted, the uncertainty caused by real-time positioning errors can be avoided, the trajectory deviation of the vehicle during driving can be effectively reduced, it can be ensured that the vehicle always travels along the planned path, and the situation of deviating from the predetermined track can be avoided, thereby improving the overall driving safety and reliability. Smoothing processing can adjust the movement trajectory of the vehicle, reduce drastic acceleration, deceleration or steering changes, make the vehicle travel more smoothly, and thus improve the comfort of passengers. Through the trajectory optimization processing, the accuracy, stability and reliability of the vehicle running trajectory are improved.

[0036] Based on the same inventive concept, the present application also proposes a system for a vehicle positioning method, and the system includes:

[0037] A data acquisition module, configured to acquire vehicle pose data and panoramic image data in real time.

[0038] An image processing module, configured to perform preliminary processing on the panoramic image data based on preset image parameters to obtain a bird's-eye view.

[0039] A target detection module, configured to detect the bird's-eye view through a target detection model to obtain target landmark data.

[0040] A target tracking module for performing real-time tracking on the target marker data based on the current map setting status.

[0041] And a pose correction module for matching the target marker data at any two consecutive moments based on a first preset matching algorithm, and correcting the vehicle pose data based on the matching result to obtain a final vehicle positioning result.

[0042] Based on the same inventive concept, the present application also provides a computer-readable storage medium storing computer-executable instructions, and when the computer-executable instructions are executed by a control processor, the vehicle positioning method is implemented.

[0043] Compared with the prior art, the present application has at least the following beneficial effects:

[0044] The vehicle positioning method provided by the present application detects the bird's-eye view through the target detection model, thereby obtaining effective target marker data, which can effectively improve the reliability of vehicle positioning. The obtained target marker data is tracked and matched in real time, thereby updating the vehicle pose data and improving the accuracy of vehicle positioning. By matching the target marker data at any two consecutive moments through the first preset matching algorithm, the problem of large computational amount of point cloud matching and difficult implementation caused by directly using semantic point clouds to construct a map in the prior art is solved, the real-time performance and stability of vehicle positioning are improved, the computational amount in the vehicle positioning process is reduced, and the system deployment cost is reduced. By performing real-time tracking on the target marker data, the robustness of the vehicle positioning result is ensured. Precise positioning correction improves the safety of autonomous driving and reduces the probability of traffic accidents. Description of the Drawings

[0045] Figure 1 is a flowchart of the vehicle positioning method shown in an embodiment of the present application.

[0046] Figure 2 is a schematic diagram of the vehicle positioning system shown in an embodiment of the present application. Detailed Embodiments

[0047] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0048] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0049] Embodiment 1:

[0050] Please refer to Figure 1 , the vehicle positioning method mainly includes steps S1 to S4.

[0051] Among them, step S1 includes: obtaining vehicle pose data and panoramic image data in real time. Among them, the vehicle pose data can be mainly collected in real time by an IMU (Inertial Measurement Unit) in combination with wheel speed pulses. The panoramic image data can be mainly obtained by shooting with four-way fish-eye cameras or panoramic cameras installed on the vehicle body.

[0052] Step S2 includes: preliminarily processing the panoramic image data based on preset image parameters to obtain a bird's-eye view. Among them, the preset image parameters can be mainly the internal parameter calibration parameters of the panoramic camera, and the preliminary processing at least includes distortion correction and inverse perspective transformation processing. That is, the panoramic image data can be corrected for distortion through the internal parameter calibration parameters of the panoramic camera, and after the distorted panoramic image data is subjected to inverse perspective transformation processing, the preliminarily processed panoramic image data is stitched into a bird's-eye view centered on the vehicle.

[0053] Step S3 includes: detecting the bird's-eye view through a target detection model to obtain target marker data. Among them, the target detection model can be a deep learning model. For example, the target detection model can be a CNN (Convolutional Neural Networks), an RNN (Recurrent Neural Networks), or a GNN (Graph Neural Networks). The target detection model can also include two parts, namely a target recognition model and a semantic segmentation model. The target marker data at least includes the target marker type, position, and edge point cloud data. The target marker type and position are recognized through the target recognition model; the bird's-eye view is semantically segmented through the semantic segmentation model, and the edge point cloud data is extracted based on the obtained semantic segmentation map. The target marker types mainly include straight arrows, turning arrows, U-turn arrows, straight-plus-turn arrows, speed bumps, lane lines, and parking space lines. Those skilled in the art can increase or decrease the target marker types according to actual situations, and are not limited thereto.

[0054] Step S4 includes: performing real-time tracking on the target marker data based on the current map setting state, and matching the target marker data at any two consecutive moments based on a first preset matching algorithm to correct the vehicle pose data based on the matching result and obtain the final vehicle positioning result. In the case where the current vehicle does not have a map set, the DeepSORT (Deep Learning-based Simple Online and Realtime Tracking) algorithm can be used to perform real-time tracking on the target marker data. The first preset matching algorithm can be the ICP (Iterative Closest Point) algorithm. Those skilled in the art can select other tracking algorithms and matching algorithms according to actual situations, and are not limited thereto.

[0055] For example, in the specific implementation process, vehicle pose data can be obtained in real time through an inertial measurement unit combined with wheel speed pulses, and panoramic image data of the vehicle can be collected in real time through panoramic cameras; based on the internal parameter calibration parameters of the panoramic cameras, the panoramic image data is subjected to distortion correction processing and inverse perspective transformation processing, and then the processed image data is stitched into a bird's-eye view; a pre-constructed and trained target detection model based on deep learning algorithms is used to detect the bird's-eye view to obtain the target landmark data; the DeepSORT algorithm is used to perform real-time tracking on the target landmark data. Based on the ICP matching algorithm, the target landmark data at any two consecutive moments is matched, and the vehicle pose data is corrected with the obtained matching result to obtain the final vehicle positioning result.

[0056] Optionally, the preliminary processing at least includes distortion correction processing and inverse perspective transformation processing; the preliminary processing of the panoramic image data includes:

[0057] Performing distortion correction processing on the panoramic image data based on the preset image parameters; performing inverse perspective transformation processing on the panoramic image data after distortion correction processing to obtain the final panoramic image data; stitching the final panoramic image data to obtain the bird's-eye view.

[0058] Among them, the preset image parameters may include focal length, distortion coefficient, etc.; the distortion correction processing mainly eliminates image distortion caused by factors such as camera lenses and imaging sensors, such as radial distortion and tangential distortion. The radial distortion is the geometric shape distortion from the center of the image outwards, causing straight lines to become curves. The tangential distortion is mainly the image offset and shape deformation caused by the camera lens not being completely parallel to the image sensor. In the specific distortion correction processing process, based on the focal length and distortion coefficient, the panoramic image data can be subjected to distortion correction through geometric transformation. The inverse perspective transformation processing mainly converts the panoramic image data into planar image data. In the process of stitching the final panoramic image data to obtain the bird's-eye view, mainly the image data can be aligned based on the position and perspective of the camera, and a suitable image stitching algorithm, such as an image fusion algorithm, can be selected to stitch the image data to obtain the bird's-eye view.

[0059] Optionally, the pre-constructed target detection model includes:

[0060] Constructing the target detection model through deep learning algorithms; obtaining historical target landmark data to train the target detection model with the historical target landmark data; performing optimization processing on the trained target detection model based on a preset optimization algorithm to obtain the final target detection model.

[0061] A large number of ground marker images under different illumination conditions, angles, and scenarios can be used to train the target detection model. The preset optimization algorithm can be an iterative algorithm. Through multiple iterative trainings and adjustments of the target detection model, the final target detection model is obtained.

[0062] Optionally, the target marker data at least includes the type, position, and edge point cloud data of the target marker; the target detection model at least includes a target recognition model and a semantic segmentation model; detecting the bird's-eye view through the target detection model to obtain target marker data includes:

[0063] Performing target recognition on the bird's-eye view through the target recognition model to obtain the type and position of the target marker.

[0064] Performing semantic segmentation on the bird's-eye view through the semantic segmentation model to extract the edge point cloud data based on the obtained semantic segmentation map.

[0065] For example, the target recognition model can be a convolutional neural network. The semantic segmentation model mainly performs pixel-level separation on the bird's-eye view. Performing target recognition on the bird's-eye view through the target recognition model to obtain the type of the target marker and its position in the bird's-eye view. Outputting a semantic segmentation map through the semantic segmentation model. After performing semantic segmentation on the bird's-eye view through the semantic segmentation model, further converting the edge contour information of the target marker extracted from the obtained semantic segmentation map into edge point cloud data.

[0066] Optionally, the vehicle pose data at least includes the first vehicle pose and the mileage timestamp, and the surround-view image data at least includes the surround-view image and the image timestamp; before performing real-time tracking on the target marker data, it further includes:

[0067] Converting the target marker data to a vehicle-centered world coordinate system.

[0068] In the world coordinate system, obtaining the mileage timestamps within a preset time threshold range based on the image timestamp.

[0069] Based on the time difference between the image timestamp and the mileage timestamp, performing a linear interpolation operation on the first vehicle pose corresponding to the mileage timestamp to obtain the second vehicle pose corresponding to the image timestamp.

[0070] Among them, the preset time threshold range can be 0.1 second or 0.2 second, and mainly the nearest mileage timestamp within the preset time threshold range of the image timestamp is obtained. Those skilled in the art can adjust the preset time threshold range according to the actual situation, and it is not limited thereto. The linear difference operation mainly can calculate the second vehicle pose corresponding to the time difference between the image timestamp and the mileage timestamp through the mileage timestamps at two consecutive moments and the corresponding first vehicle pose.

[0071] Optionally, before correcting the vehicle pose data, it includes: obtaining the current map setting state. If the current map setting state is that no map is set, then the obtaining of the final vehicle positioning result includes:

[0072] Performing real-time tracking on the target marker data at consecutive moments.

[0073] Performing preliminary matching on the target marker data at any two consecutive moments, and based on the preliminary matching result, using a first preset matching algorithm to match the edge point cloud data to obtain the translation vector of the target marker data at consecutive moments.

[0074] Constructing a vehicle pose constraint relationship at consecutive moments based on the translation vector, so as to correct the second vehicle pose based on the vehicle pose constraint relationship, and obtaining the final vehicle positioning result according to the correction result.

[0075] Among them, the deepsort algorithm can be used to perform real-time tracking on the target marker data at consecutive moments, and the first preset matching algorithm can be the ICP algorithm. For example, in the case where no map is set currently, performing real-time tracking on the target marker data at consecutive moments through the deepsort algorithm, performing preliminary matching on the target marker types and positions at any two consecutive moments, and further matching the edge point cloud data using the ICP algorithm on this basis, reducing the amount of matching operations while ensuring the matching quality. Obtaining the translation vector of the target marker data at two moments according to the matching result, constructing a pose constraint relationship of the vehicle between two adjacent moments based on the translation vector, and correcting the second vehicle pose based on the pose constraint relationship, thereby obtaining the final vehicle positioning result.

[0076] Optionally, if the current map setting state is that a map is set, the obtaining of the final vehicle positioning result includes:

[0077] Converting the target marker data to the map coordinate system to obtain the first position coordinate.

[0078] Matching the target marker data with the map through a second preset matching algorithm to obtain the second position coordinate.

[0079] Obtain a rotation and translation vector based on the first position coordinate and the second position coordinate, so as to correct the pose of the second vehicle based on the rotation and translation vector, and obtain a final vehicle positioning result according to the correction result.

[0080] Wherein, in the case where a map has been set, the first position coordinate of the target marker data in the map coordinate system can be obtained by taking the pose of the second vehicle as the starting point of the map and converting the target marker data into the map coordinate system. The second preset matching algorithm can mainly be the Hungarian matching algorithm, and the target marker data is matched with the map through the Hungarian matching algorithm, so as to obtain the second position coordinate of the target marker data.

[0081] Optionally, after obtaining the final vehicle positioning result, it further includes:

[0082] Predict the vehicle running trajectory based on the final vehicle positioning result; perform corresponding scenario processing based on the vehicle running trajectory; wherein, the scenario processing at least includes smoothing processing and trajectory optimization processing.

[0083] For example, there is a speed bump in the predicted vehicle running trajectory. The speed bump will cause the vehicle to shake and reduce the weight of visual perception in vehicle positioning. Therefore, smoothing processing is required. The time for the vehicle to pass through the speed bump can be calculated based on the vehicle speed, and smoothing processing is adopted during the process of the vehicle passing through the speed bump, and the vehicle poses before and after passing through the speed bump are recorded. When there are uphill and downhill slopes in the vehicle running trajectory, trajectory optimization processing can be adopted during the process of the vehicle passing through the uphill and downhill slopes to improve the vehicle positioning result.

[0084] Embodiment 2:

[0085] Please refer to Figure 2 , this application also proposes a system adopting the vehicle positioning method described in Embodiment 1, which mainly includes: a data acquisition module, an image processing module, a target detection module, a target tracking module, and a pose correction module.

[0086] Wherein, the data acquisition module is used to acquire vehicle pose data and panoramic image data in real time. Among them, the vehicle pose data can mainly be acquired in real time by an inertial measurement unit in cooperation with wheel speed pulses, and the panoramic image data can mainly be acquired in real time by a panoramic camera.

[0087] The image processing module is used to perform preliminary processing on the panoramic image data based on preset image parameters to obtain a bird's-eye view. The preset image parameters can mainly be the internal parameter calibration parameters of the panoramic camera, such as focal length and distortion coefficient. The preliminary processing can at least include distortion correction processing and inverse perspective transformation processing. The obtained image data after preliminary processing is stitched to obtain the bird's-eye view.

[0088] A target detection module, configured to detect the bird's-eye view through a target detection model to obtain target marker data. Wherein, the target detection model is a target detection model pre-constructed by a deep learning algorithm and trained with historical target marker data under various different working conditions, such as historical target marker data under various different illuminations, angles, and scenes to train the target detection model. The target marker data at least includes the target marker type, position, and edge point cloud data. The target detection model at least includes a target recognition model and a semantic segmentation model. The target marker type and position are recognized through the target recognition model, and the edge point cloud data of the target marker is obtained through the semantic segmentation model.

[0089] A target tracking module, configured to perform real-time tracking on the target marker data based on the current map setting state. Among them, the deepsort algorithm can be used to perform real-time tracking on the target marker data, and those skilled in the art can select other tracking algorithms according to the actual situation, which is not limited thereto.

[0090] And, a pose correction module, configured to match the target marker data at any two consecutive moments based on a first preset matching algorithm, and correct the vehicle pose data based on the matching result to obtain a final vehicle positioning result. The first preset matching algorithm can mainly be the ICP algorithm. Among them, the target marker type and position of the target marker data at any two consecutive moments are mainly initially matched, and then the edge point cloud data is further matched through the ICP algorithm. A translation vector of the target marker data at two consecutive moments is obtained based on the matching result, and then a vehicle pose constraint relationship of the vehicle at two adjacent moments is constructed based on the translation vector. The vehicle pose data is corrected based on the vehicle pose constraint relationship to obtain a final vehicle positioning result.

[0091] Embodiment Three:

[0092] The present application also proposes a domain controller, including a computer-readable storage medium, where the computer-readable storage medium includes:

[0093] The computer-readable storage medium stores computer-executable instructions.

[0094] When the computer-executable instructions are executed by a control processor, the vehicle positioning method described in Embodiment One is implemented.

[0095] In the computer-readable storage medium, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as coaxial cable, optical fiber, digital subscriber line) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state drive (SSD)).

[0096] In summary, in the present application, the bird's-eye view is detected by a pre-trained object detection model, and the obtained target landmark data is tracked and matched in real time, so as to update the vehicle pose data, improve the accuracy of vehicle positioning, and improve the accuracy and reliability of the vehicle positioning result. By using the first preset matching algorithm to match the target landmark data at any two consecutive moments, the problem of large computational complexity and difficult implementation caused by directly using semantic point clouds to construct a map in the prior art is solved, the computational complexity in the vehicle positioning process is reduced, and the system deployment cost is reduced. By tracking the target landmark data in real time, the timeliness and accuracy of the vehicle positioning result are ensured.

[0097] In several embodiments provided in the present application, it can be understood that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, the program segment, or the part of code includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the block may occur in a different order from that marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.

[0098] When the above-mentioned functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0099] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only for the specific embodiments of this application and is not used to limit the protection scope of this application. In particular, it is pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this application should be included in the protection scope of this application.

Claims

1. A vehicle positioning method, characterized in that: A target detection model is pre-built, and the vehicle positioning method includes: Acquire vehicle posture data and surround view image data in real time; Performing preliminary processing on the surround view image data based on preset image parameters to obtain a bird's-eye view; Detect the bird's-eye view image using a target detection model to obtain target marker data; The target mark data is tracked in real time based on the current map setting state, and the target mark data at any two consecutive moments are matched based on a first preset matching algorithm, so as to correct the vehicle posture data based on the matching result to obtain a final vehicle positioning result.

2. A vehicle positioning method according to claim 1, characterized in that: The preliminary processing includes at least distortion correction processing and inverse perspective transformation processing; the preliminary processing of the surround view image data includes: Performing distortion correction processing on the surround view image data based on the preset image parameters; Performing inverse perspective transformation on the surround view image data after the distortion correction process to obtain final surround view image data; The final surround image data are stitched together to obtain the bird's-eye view.

3. A vehicle positioning method according to claim 1, characterized in that: The pre-built target detection model includes: Constructing the target detection model through a deep learning algorithm; Acquiring historical target marker data to train the target detection model using the historical target marker data; The trained target detection model is optimized based on the preset optimization algorithm to obtain the final target detection model.

4. A vehicle positioning method according to claim 1, characterized in that: The target mark data at least includes the type, position and edge point cloud data of the target mark; The target detection model at least includes a target recognition model and a semantic segmentation model; the detecting the bird's-eye view through the target detection model to obtain target mark data includes: Performing target recognition on the bird's-eye view through the target recognition model to obtain the type and position of the target mark; The semantic segmentation model is used to perform semantic segmentation on the bird's-eye view image, so as to extract the edge point cloud data based on the acquired semantic segmentation image.

5. A vehicle positioning method according to claim 4, characterized in that: The vehicle posture data at least includes a first vehicle posture and a mileage timestamp, and the surround view image data at least includes a surround view image and an image timestamp; before real-time tracking of the target mark data, the method further includes: Converting the target marker data to a world coordinate system centered on the vehicle; In the world coordinate system, obtaining a mileage timestamp within a preset time threshold range based on the image timestamp; Based on the time difference between the image timestamp and the mileage timestamp, a linear difference operation is performed on the first vehicle posture corresponding to the mileage timestamp to obtain a second vehicle posture corresponding to the image timestamp.

6. A vehicle positioning method according to claim 5, characterized in that: Before correcting the vehicle posture data, the process includes: obtaining the current map setting state. If the current map setting state is that the map is not set, obtaining the final vehicle positioning result includes: Tracking the target marker data at consecutive moments in real time; Preliminary matching is performed on the target mark data at any two consecutive moments, and based on the preliminary matching result, the edge point cloud data is matched using a first preset matching algorithm to obtain the translation vector of the target mark data at consecutive moments; A vehicle posture constraint relationship at consecutive moments is constructed based on the translation vector, so as to correct the second vehicle posture based on the vehicle posture constraint relationship, and obtain a final vehicle positioning result according to the correction result.

7. A vehicle positioning method according to claim 5, characterized in that: If the current map setting state is that the map has been set, the obtaining of the final vehicle positioning result includes: Converting the target mark data into a map coordinate system to obtain a first position coordinate; Matching the target mark data with the map by a second preset matching algorithm to obtain second position coordinates; A rotational translation vector is obtained based on the first position coordinate and the second position coordinate, so as to correct the second vehicle posture based on the rotational translation vector, and a final vehicle positioning result is obtained according to the correction result.

8. A vehicle positioning method according to claim 1, characterized in that: After obtaining the final vehicle positioning result, the method further includes: Predicting a vehicle running trajectory based on the final vehicle positioning result; Taking corresponding scene processing based on the vehicle running trajectory; The scene processing includes at least smoothing processing and trajectory optimization processing.

9. A system based on the vehicle positioning method according to any one of claims 1 to 8, characterized in that: The system comprises: Data acquisition module, used to acquire vehicle posture data and surround view image data in real time; An image processing module, used to perform preliminary processing on the surround view image data based on preset image parameters to obtain a bird's-eye view; a target detection module, used to detect the bird's-eye view through a target detection model to obtain target marker data; A target tracking module, used for tracking the target mark data in real time based on the current map setting state; And, a posture correction module is used to match the target mark data at any two consecutive moments based on a first preset matching algorithm, so as to correct the vehicle posture data based on the matching result to obtain a final vehicle positioning result.

10. A computer-readable storage medium storing computer-executable instructions, characterized in that: When the computer executable instructions are executed by the control processor, a vehicle positioning method as described in any one of claims 1-8 is implemented.

Images