A vehicle positioning method, electronic device and system based on low computing power platform

Abstract

The present application provides a vehicle positioning method, electronic device and system based on a low computing power platform. The vehicle positioning method includes: obtaining map construction information according to the current frame information of the vehicle; matching the map construction information with the preset static map to obtain a local map; based on the camera coordinate system, projecting the point cloud of the local map to the image data, and obtaining a camera photo containing depth information as depth image information; using the segmentation model of the low computing power platform to perform semantic segmentation of ground markings on the depth image information, using the segmented point cloud as the ground marking point cloud, and relocating the vehicle according to the ground marking point cloud; the vehicle includes at least one of an automatic inspection vehicle and an energy storage charging vehicle, the target area is an area where real-time dynamic measurement signals are missing, and the segmentation model is a UNet network model. It can be used to achieve vehicle positioning through a low computing power platform in target areas such as underground parking lots where real-time dynamic measurement signals are missing.

Description

Technical Field

[0001] The present application relates to the field of vehicle positioning technology, and in particular to a vehicle positioning method, electronic equipment and system based on a low computing power platform. Background Art

[0002] Lane marking recognition in the field of autonomous driving is achieved through traditional vision-based methods (such as using the OpenCV library) and can be used for vehicle positioning. However, lane marking recognition based on traditional vision mainly relies on feature rules such as color, shape or texture to identify lane markings, etc. Its disadvantage is that it is not robust and is easily affected by environmental factors such as lighting and weather. It also lacks depth information, which limits its ability to accurately locate lane markings in three-dimensional space.

[0003] Based on this, a method of lane marking recognition based on deep learning is proposed in the relevant technology, that is, using deep neural networks to recognize lane markings and handle more complex scenes and lighting conditions. This method requires a lot of computing power support, which is difficult to implement for vehicles with low computing power platforms (such as low-power embedded systems) with limited computing resources. This is because vehicles with low computing power platforms can only be equipped with processors with low performance and cannot run computationally intensive algorithms; the battery capacity is also limited and cannot support high-computing power calculations.

[0004] Therefore, how to achieve vehicle positioning based on a low computing power platform is one of the technical problems that currently need to be solved in this field. Summary of the invention

[0005] The purpose of this application is to provide a vehicle positioning method, electronic equipment and system based on a low computing power platform, which solves the problem of vehicle positioning on a low computing power platform with limited computing resources.

[0006] The purpose of this application is achieved by the following technical solutions:

[0007] In a first aspect, the present application provides a vehicle positioning method based on a low computing power platform, which is used for positioning a vehicle in a target area, comprising the following steps:

[0008] Acquiring map construction information according to current frame information of the vehicle; the current frame information includes image data, laser radar data and inertial measurement data;

[0009] Matching the map construction information with a preset static map to obtain a local map;

[0010] Based on a camera coordinate system, projecting the point cloud of the local map onto the image data, and obtaining a camera photo containing depth information as depth image information;

[0011] Using a segmentation model of a low-computing-power platform to perform semantic segmentation of ground markings on the depth image information, using the segmented point cloud with depth information as a ground marking point cloud, and relocating the vehicle according to the ground marking point cloud;

[0012] Wherein, the vehicle includes at least one of an automatic inspection vehicle and an energy storage charging vehicle, the target area is an area where real-time dynamic measurement signals are missing, and the segmentation model is a UNet network model.

[0013] The beneficial effect of the technical solution is that in the vehicle positioning method, the current frame information refers to the environmental data captured by the vehicle's sensor, reflecting the instantaneous state of the vehicle's surrounding environment at that moment. The camera module, laser measurement module and inertial measurement module can respectively collect image data, lidar data and inertial measurement data and together constitute the complete information of the current frame for subsequent processing and analysis. Then, the lidar data is dedistorted using the inertial measurement data (IMU) to correct the data distortion caused by sensor errors, thereby obtaining accurate map construction information. Dynamic objects (such as vehicles, pedestrians, etc.) in the lidar data are identified and removed by an algorithm to obtain purified lidar data in preparation for subsequent precise positioning. The map construction information is matched with a pre-built preset static map, and a local map of the preset static map is obtained after the match is successful. The point cloud data of the local map is projected into the camera coordinate system, and depth information is assigned to each pixel in the image data, thereby generating a camera photo containing depth information, i.e., depth image information. Pixels without depth values ​​do not participate in subsequent feature operations to improve processing efficiency. The ground markings are segmented and extracted from the depth image information, and the ground markings are identified using image segmentation technology (such as UNet network). The extracted ground marking point cloud will be used to update the point cloud information of the local map. The updated local map point cloud information will be used for vehicle relocation. It can be considered that during the relocation process, the local map point cloud information can be used to accurately determine the position of the vehicle in the map.

[0014] Compared with the related art, a method based on deep learning is required to use a trained multi-task deep neural network to process the continuous image sequence to be tested, and obtain the pose transformation information, depth information and semantic segmentation information between image frames, which requires relatively high computing power. This application aims to meet the challenges of high-dynamic environments in areas where RTK signals are easily lost. After adding the necessary semantic information extracted by UNet segmentation, the applicability and robustness of the vehicle positioning method can be greatly improved. The above-mentioned vehicle positioning method does not require an additional CUDA graphics card, which saves power and economic costs.

[0015] Furthermore, acquiring map construction information according to the current frame information of the vehicle includes:

[0016] The laser radar data is subjected to dedistortion processing according to the inertial measurement data, and dynamic objects are removed from the laser radar data after the distortion processing to obtain map construction information.

[0017] The beneficial effect of this technical solution is that, due to sensor errors, vehicle motion or environmental factors, the lidar data contains distortion, and de-distortion processing is performed based on the inertial measurement data to correct the error, so that the map construction information obtained is more accurate.

[0018] Furthermore, before obtaining the map construction information, the method further includes:

[0019] Calibrate the external parameters of the camera module, laser measurement module and inertial measurement module so that their mutual relations are fixed parameters;

[0020] The camera module is used to acquire image data of the vehicle, the laser measurement module is used to acquire laser radar data of the vehicle, and the inertial measurement module is used to acquire inertial measurement data of the vehicle; and the image data, the laser radar data and the inertial measurement data are used as current frame information.

[0021] The beneficial effect of this technical solution is that resources are limited on low-computing power platforms. This technical solution ensures that sensor data is processed efficiently through external parameter calibration (and possible subsequent data fusion), avoiding unnecessary computing burden, thereby optimizing the utilization of limited resources, and further reducing computing-intensive tasks such as deep learning reasoning, which helps to reduce overall energy consumption and extend battery life.

[0022] Furthermore, the process of projecting the point cloud in the local map onto the image data includes:

[0023] Performing posture interpolation according to respective timestamps of the image data and the laser radar data to determine a specific posture of the camera module in the local map;

[0024] According to the internal parameters of the camera module, the point cloud data in the local map is projected onto a two-dimensional image plane corresponding to the image data, and a camera photo containing depth information is generated as depth image information.

[0025] The beneficial effect of this technical solution is that the depth image information provides the vehicle with richer environmental perception data. The data obtained through the subsequent segmentation process enables the vehicle to better understand its surrounding environment, including the location of obstacles, road edges, and ground markings. By removing dynamic objects and using static features (such as ground markings), this technical solution improves robustness in dynamic environments and reduces positioning errors caused by environmental changes.

[0026] In summary, this technical solution generates depth image information containing depth information through precise pose interpolation and point cloud projection, thereby improving the vehicle's positioning accuracy and environmental perception capabilities, while reducing dependence on high-performance hardware, and has significant economic benefits and application value.

[0027] Furthermore, the point cloud of the local map only includes the laser point cloud of the ground, and the projecting of the point cloud data in the local map onto a two-dimensional image plane corresponding to the image data according to the intrinsic parameters of the camera module includes:

[0028] For each laser point cloud, first convert it according to the following formula:

[0029]

[0030] Then the transformed points are dedistorted to realize the projection of point cloud data. The dedistortion process corresponds to the following formula:

[0031]

[0032] Among them, K is the camera internal parameter matrix, the coefficient (( k 1, k 2, k 3) is the radial distortion, the coefficient ( p 1, p 2) is the tangential distortion, x and y are the coordinates of the laser point cloud, x corrected , y corrected Corresponding to the data after x and y distortion, r 2 is the normalized distance from the laser point cloud to the image center.

[0033] The beneficial effect of this technical solution is that, first, for each laser point cloud in the local map, coordinate transformation is performed according to the intrinsic parameter K of the camera module. K includes the focal length and optical center coordinates of the camera, which are used to transform the three-dimensional world coordinates into two-dimensional image coordinates in the camera coordinate system. The transformed point cloud data needs to be dedistorted, and the dedistorted point cloud data is projected onto a two-dimensional image plane corresponding to the image data to generate a camera photo containing depth information.

[0034] In summary, through accurate internal parameter matrix and de-distortion processing, the conversion accuracy of laser point cloud to image coordinates is improved under the limited computing power of low computing power platform, thereby improving the positioning accuracy of the vehicle. The de-distorted point cloud data can more realistically reflect the environmental conditions and enhance the ability to perceive the environment.

[0035] Furthermore, the method further comprises: determining a current working mode of the vehicle, the current working mode comprising a positioning mode and a mapping mode;

[0036] When the working mode is the positioning mode, repositioning the vehicle according to the ground marking point cloud includes:

[0037] Loading the historical key point cloud to form a local historical map with only ground markings; matching the ground marking point cloud with the local historical map, and outputting the repositioned posture of the vehicle;

[0038] When the working mode is the mapping mode, the ground marking point cloud is associated with the current position of the vehicle and saved as a ground marking key frame.

[0039] The beneficial effect of this technical solution is that by using historical key point clouds and local historical maps, the vehicle's position can be determined more accurately in environments where the RTK signal is poor or missing. The historical key point cloud data is effectively reused, reducing the dependence on real-time sensor data, improving the efficiency of data processing and the robustness of the system. By distinguishing between positioning mode and mapping mode, the working mode can be flexibly switched according to actual needs. For applications such as automatic inspection vehicles and energy storage charging vehicles, accurate positioning and effective mapping can improve work efficiency, reduce manual intervention, and improve user experience.

[0040] Furthermore, the low computing power platform is built on a vehicle controller of the vehicle.

[0041] The beneficial effect of this technical solution is that it utilizes the vehicle's existing vehicle controller (VCU), reducing the need for additional high-performance computing hardware, thereby reducing the cost of the system. Since the VCU is usually designed as a low-power device, it helps with the energy consumption of automatic inspection vehicles or energy storage charging vehicles and increases the battery life. Integrating data processing and decision-making in the VCU simplifies the electronic architecture of the vehicle. Centralized processing reduces the need for data transmission between multiple computing units, reduces the risk caused by communication errors, and improves the reliability of the vehicle positioning system.

[0042] In a second aspect, the present application further provides an electronic device, wherein the electronic device comprises a memory and a processor, the memory stores a computer program, and the processor is configured to implement the steps of any of the above-described methods when executing the computer program.

[0043] In a third aspect, the present application also provides a vehicle positioning system, comprising the electronic device described in the second aspect, and also comprising an information acquisition device electrically connected to the electronic device; the information acquisition device comprises a camera module, a laser measurement module and an inertial measurement module, the camera module is used to acquire image data, the laser measurement module is used to acquire lidar data, and the inertial measurement module is used to acquire inertial measurement data. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present application is further described below in conjunction with the accompanying drawings and embodiments.

[0045] Figure 1 It is a flow chart of a vehicle positioning method provided in an embodiment of the present application.

[0046] Figure 2 It is a partial flow chart of a vehicle positioning method provided in an embodiment of the present application.

[0047] Figure 3 It is a flow chart of another vehicle positioning method provided in an embodiment of the present application.

[0048] Figure 4 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] Below, in conjunction with the accompanying drawings and specific implementation methods, the present application is further described. It should be noted that, under the premise of no conflict, the various embodiments described below or the various technical features can be arbitrarily combined to form a new embodiment. The implementation procedures of the present application will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific implementation procedures, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be understood that the preferred embodiments are only for illustrating the present application, not for limiting the scope of protection of the present application.

[0050] Below, some terms in this application are explained to facilitate understanding by those skilled in the art.

[0051] RTK (Real-Time Kinematic) positioning is a high-precision positioning technology based on GPS (Global Positioning System) or GNSS (Global Navigation Satellite System), which can provide real-time, centimeter-level positioning accuracy. However, RTK (Real-Time Kinematic Measurement) signals are easily affected by environmental factors (such as obstructions such as buildings and trees), which in turn affects the positioning of the vehicle during autonomous driving.

[0052] Static maps (preset static maps) are pre-built maps that contain static features in the environment. Static features are features that do not change over time. In specific applications, static maps are usually used for autonomous driving, robot navigation, geographic information systems (GIS), and other applications that require location information.

[0053] The physical model of the UNet architecture is a convolutional neural network (CNN) model designed for image segmentation tasks, which can fuse deep contextual information of images with precise location information, thereby providing excellent performance when performing fine segmentation tasks. The structure of UNet consists of two parts: an encoder (contracting path) that captures image features and gradually reduces spatial dimensions; and a decoder (expanding path) that gradually restores the spatial resolution of the image. The encoder and decoder are bridged by skip connections, directly connecting the high-resolution feature map in the encoder with the corresponding low-resolution feature map in the decoder, thereby retaining rich location information during the decoding process of the network.

[0054] In the related technology, the vehicle positioning can be achieved through the traditional vision (openCV) detection technology. Taking the parking lot path finding solution based on image recognition and positioning disclosed in application number CN202410873452.4 as an example, by storing the image data of the parking lot in the data processing center and performing image recognition to obtain the location information based on the image information of the parking space taken by the user, the positioning of the vehicle and the user and the generation of the vehicle path can be achieved, which has the advantage of low cost. However, its technical solution is more rule-based, resulting in weak robustness of the positioning result. In addition, the information used for positioning lacks depth information. The depth information provides the distance between the object and the camera. The lack of depth information will lead to reduced positioning accuracy.

[0055] In the related art, in order to provide depth information, taking application number CN201810236640.0 as an example, an end-to-end semantic instant positioning and mapping method based on deep learning is disclosed. A continuous original image sequence and an original three-dimensional point cloud sequence corresponding to the image sequence are collected by a color camera and a laser radar respectively, and the pose transformation information, depth information and semantic segmentation information of a continuous five-frame image sequence are obtained by processing; a multi-task deep neural network with branches is constructed, input into the multi-task deep neural network, and the multi-task deep neural network is trained to obtain parameters. The trained multi-task deep neural network is used to process the five-frame image sequence to be tested, and the pose transformation information, depth information and semantic segmentation information between the image frames are obtained. Compared with the traditional ORB-SLAM algorithm and the method based on deep learning, it has better performance. However, while providing better performance, deep learning still requires a lot of computing power. For vehicles such as automatic inspection vehicles and energy storage charging vehicles mentioned in this application, it means that the computing power of the vehicle itself is relatively low and the work intensity is relatively high. The semantic instant positioning and mapping methods mentioned above require additional hardware support, such as high-performance graphics processing units (GPUs) or dedicated artificial intelligence processors to increase computing power, which are relatively costly and power-intensive.

[0056] However, the mobile platforms of working vehicles such as automatic inspection vehicles and energy storage charging vehicles are sensitive to cost and power consumption, and cannot have large-capacity batteries and processors with large computing power like cars in the consumer market. Automatic inspection vehicles and energy storage charging vehicles are used in underground parking lots and other places in high-dynamic environments. Different vehicles and different parking make the surrounding environment information change greatly over time, which makes it difficult to complete the task of laser matching repositioning. Based on this, the present application provides a vehicle positioning method, electronic equipment and system based on a low-computing power platform. In the vehicle positioning method protected by the present application, considering that the ground markings (including the lane markings and road markings mentioned below) are very stable features and will not change over time, it can provide relatively stable features in the ever-changing parking environment, achieve relatively stable repositioning, and then provide another repositioning observation for the long-term operation of the laser odometer. Therefore, it can be used to locate the vehicle through a low-computing power platform in target areas such as underground parking lots where real-time dynamic measurement signals are missing. The following will first describe the vehicle positioning method, and then describe the electronic equipment and vehicle positioning system.

[0057] Method embodiment.

[0058] See also Figure 1 , Figure 1 It is a flow chart of a vehicle positioning method provided in an embodiment of the present application.

[0059] This embodiment provides a vehicle positioning method based on a low computing power platform, wherein the vehicle includes at least one of an automatic inspection vehicle and an energy storage charging vehicle, the target area in the vehicle positioning method is an area where real-time dynamic measurement signals are missing, and the segmentation model is a UNet network model. The vehicle positioning method includes S101 to S105.

[0060] S101, acquiring map construction information according to current frame information of the vehicle; the current frame information includes image data, lidar data and inertial measurement data.

[0061] In a specific application, the laser radar data may be dedistorted according to the inertial measurement data, and dynamic objects may be removed from the distorted laser radar data to obtain map construction information.

[0062] Due to sensor errors, vehicle motion or environmental factors, the laser radar data contains distortion. This embodiment performs de-distortion processing based on the inertial measurement data to correct the error, so that the map construction information obtained is more accurate. The process of performing de-distortion processing based on the inertial measurement data is, for example:

[0063] Alignment through timestamps ensures that lidar data and inertial measurement data are synchronized in time;

[0064] Obtain the distortion model of the LiDAR data. The distortion model is used to describe and quantify the errors in the point cloud data caused by vehicle motion (such as acceleration, turning, etc.);

[0065] The coordinates of each laser point are converted to the global coordinate system of the vehicle, and then according to the vehicle's motion state and distortion model, the inverse kinematics transformation is applied to correct the error caused by the vehicle's motion, adjust the position of each point, and eliminate the distortion caused by motion. The distorted point cloud data is used to update the lidar data.

[0066] The dynamic objects mentioned in this application refer to objects that can move relative to places such as parking lots, such as pedestrians and vehicles. It can be considered that dynamic objects will affect the map construction, resulting in inaccurate generated maps. Removing dynamic objects ensures that only static features are included in the map, thereby improving the stability and reliability of the map.

[0067] S102: Match the map construction information with a preset static map to obtain a local map.

[0068] This application uses a preset static map because the preset static map will not be affected by the lack or instability of the RTK positioning signal in the scenarios of vehicle autonomous driving and operation (underground garages, etc.). It can be considered that in the process of acquiring the current frame information, the sensor data (such as lidar data) will be matched with the known map information to determine the approximate position and direction of the vehicle. In this step, the point cloud of the moving object will be removed and used for matching again to obtain a more accurate posture. According to the newly acquired posture, the point cloud over a certain time and space is accumulated as a local map (subMap) of the preset static map to achieve the creation of a local (point cloud) map.

[0069] S103, projecting the point cloud of the local map onto image data based on a camera coordinate system, and acquiring a camera photo containing depth information as depth image information.

[0070] That is to say, the point cloud in the local map is projected onto the image data, and depth values ​​are assigned to some pixels in the 2D image obtained by the camera corresponding to the point cloud. In theory, images without depth values ​​do not participate in feature calculations.

[0071] The process of projecting the point cloud in the local map to the image data includes:

[0072] Performing posture interpolation according to respective timestamps of the image data and the laser radar data to determine a specific posture of the camera module in the local map;

[0073] According to the internal parameters of the camera module, the point cloud data in the local map is projected onto a two-dimensional image plane corresponding to the image data, and a camera photo containing depth information is generated as depth image information.

[0074] It can be considered that by applying the camera's intrinsic matrix, the three-dimensional point cloud data in the local map is converted into pixel coordinates in the two-dimensional image coordinate system. If the same position where the point cloud data and the image data overlap indicates a ground line, the corresponding point is considered to be a ground line.

[0075] S104, using a segmentation model of a low computing power platform to perform semantic segmentation of ground markings on the depth image information, using the segmented point cloud with depth information as a ground marking point cloud, and relocating the vehicle based on the ground marking point cloud.

[0076] Extract the point cloud of ground markings as the ground marking point cloud of the current key point cloud, that is, use the point cloud of the image with depth information as the ground marking point cloud. In this step, the image is semantically segmented for ground markings, and the part of the image with point cloud projection (i.e., the lane line part) after segmentation is marked as semantic information and used for subsequent mapping or relocalization, rather than full semantic segmentation.

[0077] In summary, this embodiment provides a vehicle positioning method, and the current frame information refers to the environmental data captured by the vehicle's sensor, which reflects the instantaneous state of the vehicle's surrounding environment at that moment. The camera module, laser measurement module and inertial measurement module can respectively collect image data, laser radar data and inertial measurement data, and together constitute the complete information of the current frame for subsequent processing and analysis. Then, the laser radar data is dedistorted using the inertial measurement data (IMU) to correct the data distortion caused by the sensor error, thereby obtaining accurate map construction information. Dynamic objects (such as vehicles, pedestrians, etc.) in the laser radar data are identified and removed by an algorithm to obtain purified laser radar data to prepare for subsequent precise positioning. The map construction information is matched with a pre-built preset static map, and a local map of the preset static map is obtained after the match is successful. The point cloud data of the local map is projected into the camera coordinate system, and depth information is assigned to each pixel in the image data, thereby generating a camera photo containing depth information, that is, depth image information. Pixels without depth values ​​do not participate in subsequent feature operations to improve processing efficiency. The ground markings are segmented and extracted from the depth image information, and the ground markings are identified using image segmentation technology (such as UNet network). The extracted ground marking point cloud will be used to update the point cloud information of the local map. The updated local map point cloud information will be used for vehicle relocation. It can be considered that during the relocation process, the local map point cloud information can be used to accurately determine the position of the vehicle in the map.

[0078] Compared with the related art, a method based on deep learning is needed to process the continuous image sequence to be tested using a trained multi-task deep neural network to obtain the pose transformation information, depth information and semantic segmentation information between the image frames, which requires relatively high computing power. The technical solution provided in this embodiment, in view of the challenge of high dynamic environments in areas where RTK signals are easily missing, adds the necessary semantic information extracted by UNet segmentation, and can greatly improve the applicability and robustness of the vehicle positioning method in a specific high dynamic environment. The method for implementing the above-mentioned vehicle positioning does not require an additional cuda graphics card, which saves power and economic costs. Because the specific semantic information in the specific scenario provided by this application is fixed, UNet has very great advantages in real-time performance and the small scale of the algorithm itself in the specific scenario provided by this application. In combination with the actual situation that maintaining real-time performance on a low computing power platform is a challenge, the above-mentioned method of optimizing the data processing process and reducing the computing requirements ensures that it can respond quickly to environmental changes and meet the needs of real-time positioning of vehicles in areas where real-time dynamic measurement signals are missing.

[0079] As an example, firstly, a joint calibration of the camera module and the laser measurement module (camera-Lidar joint calibration) is performed.

[0080] The lvt2calib calibration algorithm can be used to pre-calibrate the parameters of the camera and 3D Lidar to obtain the camera intrinsic parameter matrix K and distortion parameter D respectively. The distortion parameter includes radial distortion data and tangential distortion data, which are used to describe and correct image distortion caused by imperfect lens shape.

[0081] Among them, the internal parameter matrix K is expressed as:

[0082]

[0083] and are the focal lengths in the x-axis and y-axis directions (in pixels), and are the coordinates of the optical center (center of the image).

[0084] Radial distortion is measured by the coefficient ( , , ,...), etc., radial distortion makes straight lines in the image appear as curves, especially those near the edge of the image.

[0085] Tangential distortion is expressed by the coefficient ( , ) etc. Tangential distortion occurs because the lens is not perfectly parallel to the image sensor, causing some areas of the image to appear closer or farther away than they actually are.

[0086] The distortion parameter D can be expressed as:

[0087]

[0088] For point , The distortion correction of , its basic mathematical form can be expressed as follows:

[0089] in r 2 =x 2 +y 2 is the normalized distance from the point to the center of the image.

[0090] And the transformation matrix of the rotation matrix R and the translation vector T:

[0091]

[0092] The rotation matrix R is The translation vector T is

[0093] The combination is written as a transformation matrix:

[0094] vector is the point coordinate in the physical world, that is, the 3D point cloud world coordinate we want to use, and S is the scaling factor. Since it is a real-world projection model, its value is 1. Thus, the joint calibration of camera-Lidar (camera module and laser measurement module) is realized.

[0095] Then, according to the time point of the camera taking the photo, the coordinate system of the point cloud (multi-frame data fusion) is transformed in the laser radar coordinate system. The camera and the laser radar can be synchronized in time. Specifically, the ptp protocol can be used to synchronize the clock with the host; if there is GNSS, the host can be synchronized at the same time to achieve the purpose of time synchronization.

[0096] Calculate this moment based on the camera's shutter (global shutter) shooting time Global pose of LiDAR ; Based on the historical position of the radar, find the closest lidar position that is less than or equal to its time ;according to time, interpolate At this moment, the IMU test value is based on the time ratio , and interpolate:

[0097]

[0098] The acceleration interpolation is:

[0099]

[0100] The angular velocity interpolation is:

[0101]

[0102] Laser Based on, the following IMU pre-integration is performed to calculate its Position of:

[0103]

[0104]

[0105]

[0106]

[0107] Convert the global point cloud to get this moment Global pose of LiDAR :

[0108]

[0109] In the related technology, even if UNet can be used to identify road markings, the features of the markings it identifies do not only include those on the ground. Since it is not a 3D image, some negative features (markings on non-ground roads) will also be identified as ground markings.

[0110] Then, the point cloud of the local map only contains the laser point cloud of the ground, and the process of projecting the point cloud of the local map to the image data (i.e., reprojecting the Lidar ground point cloud to the camera coordinate system) includes:

[0111] For each laser point cloud (this laser point cloud only contains the laser point cloud on the ground, in order to intercept the data within a certain camera field of view, thereby accelerating the calculation), first convert it

[0112]

[0113] Then perform dedistortion, the dedistortion process corresponds to the following formula:

[0114] Among them, K is the camera internal parameter matrix, the coefficient ( k 1, k 2, k 3) is the radial distortion (coefficient), the coefficient ( p 1, p 2) is the tangential distortion (coefficient), x and y are the coordinates of the laser point cloud, x corrected , y corrected Corresponding to the data after x and y distortion, r 2 Normalized distance from the laser point cloud to the image center.

[0115] In this case, the three-dimensional point cloud data in the local map is converted into pixel coordinates in the two-dimensional image coordinate system by applying the camera's intrinsic parameter matrix. If the same position where the point cloud data and the image data overlap indicates a ground line, the corresponding point is considered to be a ground line that does not include areas outside the ground.

[0116] That is to say, first, for each laser point cloud in the local map, coordinate transformation is performed according to the intrinsic parameter K of the camera module. K contains the focal length and optical center coordinates of the camera, which is used to transform the three-dimensional world coordinates into two-dimensional image coordinates in the camera coordinate system. The transformed point cloud data needs to be dedistorted, and the dedistorted point cloud data is projected onto the two-dimensional image plane corresponding to the image data to generate a camera photo containing depth information.

[0117] Through precise internal parameter matrix and de-distortion processing, the conversion accuracy of laser point cloud to image coordinates is improved under the limited computing power of low computing power platforms, thereby improving the positioning accuracy of vehicles. The de-distorted point cloud data can more realistically reflect the environmental conditions and enhance the ability to perceive the environment.

[0118] Finally, the relevant point clouds are stitched into ground marking point clouds. Specifically, the point clouds identified as road markings are stitched, and the repeated dense parts are removed by downsampling (for example, the point cloud is divided into a three-dimensional voxel grid, and the points in each voxel are replaced by their center points or average points, so that adjacent points are merged, thereby reducing the number of points), to obtain laser-based road markings.

[0119] The segmentation model in this embodiment is obtained by pre-training, and the training process is, for example:

[0120] Obtain an image dataset for training the model, the image dataset containing images and their corresponding annotations. Specifically, the image is preprocessed such as scaling, normalization, and enhancement to improve the generalization ability and training efficiency of the model. At the same time, ensure that each training image has a corresponding annotation, and the annotation defines the category of each pixel in the image. During training, input the image in the image dataset, forward propagate through the U-Net model, and generate a segmentation map. Calculate the value of the loss function based on the predicted segmentation map and the true annotation. Calculate the gradient based on the loss function, and update the weight of the model through back propagation. Repeat the process of forward propagation, loss calculation, and back propagation until the performance of the model on the validation set is no longer significantly improved or reaches a predetermined number of iterations. The number of iterations is, for example, 1000, 1200, 3000, etc. Use independent validation sets and test sets to evaluate the performance of the model, and adjust the hyperparameters based on the performance of the model on the validation set. Export the trained model as the segmentation model in this embodiment.

[0121] See also Figure 2 , Figure 2 It is a partial flow chart of a vehicle positioning method provided in an embodiment of the present application.

[0122] In one embodiment, before obtaining the map construction information, the method further includes:

[0123] S105, calibrating the external parameters of the camera module, the laser measurement module and the inertial measurement module so that the relationships between them are fixed parameters;

[0124] S106, using the camera module to acquire image data of the vehicle, using the laser measurement module to acquire laser radar data of the vehicle, and using the inertial measurement module to acquire inertial measurement data of the vehicle;

[0125] S107, using the image data, the laser radar data and the inertial measurement data as current frame information.

[0126] The steps to obtain the current frame information of the vehicle involve collecting image data, laser radar data and inertial measurement data through the camera module (camera), laser measurement module (Lidar) and inertial measurement module (IMU). Among them, image data is a two-dimensional image captured by the camera, which contains visual information of the vehicle's surrounding environment, such as road signs, traffic signals, other vehicles and pedestrians. Lidar data is three-dimensional point cloud data captured by the laser radar. Inertial measurement data is three-dimensional acceleration and angular velocity data provided by the inertial measurement module (IMU), which is used to measure the dynamic movement of the vehicle including speed, acceleration and attitude changes.

[0127] In specific applications, data from different sensors may be captured at different time points, so it is necessary to synchronize the data in the current frame information to ensure data consistency. Specifically, each image data, lidar data, and inertial measurement data includes a timestamp, and the image data, lidar data, and inertial measurement data can be aligned by the timestamp. In this case, the current frame information can be understood as the information after the image data, lidar data, and inertial measurement data are aligned according to their respective timestamp data.

[0128] Therefore, resources are limited on low-computing power platforms. This technical solution ensures that sensor data (image data, lidar data, and inertial measurement data) are efficiently processed through external parameter calibration (and possible subsequent data fusion), avoiding unnecessary computing burden, thereby optimizing the utilization of limited resources and reducing computing-intensive tasks such as deep learning reasoning, which helps to reduce overall energy consumption and extend battery life.

[0129] In one embodiment, the process of projecting the point cloud in the local map to the image data includes:

[0130] Performing posture interpolation according to respective timestamps of the image data and the laser radar data to determine a specific posture of the camera module in the local map;

[0131] According to the internal parameters of the camera module, the point cloud data in the local map is projected onto a two-dimensional image plane corresponding to the image data, and a camera photo containing depth information is generated as depth image information.

[0132] In this embodiment, data from the camera and lidar with their respective timestamps are first collected. The timestamps are used to perform pose interpolation, which is used to estimate the precise position and pose of the vehicle. Since the external parameters were calibrated in the previous step, the correct alignment between different sensor data can be ensured. Once the pose of the camera in the local map is determined, the three-dimensional point cloud data in the local map can be projected onto a two-dimensional image plane using the camera's intrinsic parameters. Specifically, the three-dimensional space coordinates can be converted into two-dimensional pixel coordinates while retaining the depth information of each pixel. In this way, the generated image contains not only visual information, but also the depth information of each pixel relative to the camera.

[0133] The projected point cloud data is combined with the original image data to form depth image information. In the depth image information, each pixel also contains its corresponding depth value, thus providing the necessary three-dimensional information for subsequent image processing and analysis.

[0134] As a result, the depth image information provides the vehicle with richer environmental perception data. The data obtained through the subsequent segmentation process enables the vehicle to better understand its surroundings, including the location of obstacles, road edges, and ground markings. By removing dynamic objects and using static features (such as ground markings), this technical solution improves robustness in dynamic environments and reduces positioning errors caused by environmental changes.

[0135] In summary, this technical solution generates depth image information containing depth information through precise pose interpolation and point cloud projection, thereby improving the vehicle's positioning accuracy and environmental perception capabilities, while reducing dependence on high-performance hardware, and has significant economic benefits and application value.

[0136] In one embodiment, the method further includes: determining a current working mode of the vehicle, the current working mode including a positioning mode and a mapping mode;

[0137] When the working mode is the positioning mode, repositioning the vehicle according to the ground marking point cloud includes:

[0138] Loading the historical key point cloud to form a local historical map with only ground markings; matching the ground marking point cloud with the local historical map, and outputting the repositioned posture of the vehicle;

[0139] When the working mode is the mapping mode, the ground marking point cloud is associated with the current position of the vehicle and saved as a ground marking key frame.

[0140] In this embodiment, the current working mode of the vehicle is first determined, which can be based on user input or system settings, and then different paths of subsequent processing procedures are determined. In positioning mode, the historical key point cloud is loaded from (the storage medium), and it can be considered that the loaded historical key point cloud contains the ground marking information previously captured in the same or similar environment. Using the loaded historical key point cloud, a local historical map containing ground markings is constructed. The local historical map reflects the location and environmental characteristics of the vehicle at historical moments. The currently captured ground marking point cloud is matched with the local historical map to determine the precise position and posture of the vehicle in the map (through a feature matching algorithm), thereby achieving repositioning.

[0141] In the mapping mode, the currently captured ground marking point cloud is associated with the current position of the vehicle to ensure that the point cloud data is consistent with the actual position of the vehicle. The associated ground marking point cloud is saved as a ground marking keyframe, which will be used for subsequent map construction and updates.

[0142] Therefore, by using historical key point clouds and local historical maps, the vehicle's position can be determined more accurately in environments where the RTK signal is poor or missing. The historical key point cloud data is effectively reused, reducing the reliance on real-time sensor data, improving the efficiency of data processing and the robustness of the system. By distinguishing between positioning mode and mapping mode, the working mode can be flexibly switched according to actual needs. For applications such as automatic inspection vehicles and energy storage charging vehicles, accurate positioning and effective mapping can improve work efficiency, reduce manual intervention, and improve user experience.

[0143] In one embodiment, the low computing power platform is built on a vehicle control unit (VCU) of the vehicle.

[0144] The VCU collects current frame information by connecting to various sensors on the vehicle (such as cameras, lidar, and IMU). The VCU pre-processes and analyzes the collected information. As a result, the existing VCU of the vehicle is utilized, reducing the need for additional high-performance computing hardware, thereby reducing the cost of the system. Since the VCU is usually designed as a low-power device, it helps with the energy consumption of automatic inspection vehicles or energy storage charging vehicles and increases the battery life. Integrating data processing and decision-making in the VCU simplifies the electronic architecture of the vehicle. Centralized processing reduces the need for data transmission between multiple computing units, reduces the risk caused by communication errors, and improves the reliability of the vehicle positioning system.

[0145] In one embodiment, the method further includes: determining whether the current frame information is the first frame, and if it is the first frame, taking the current frame information as reference frame information and adding it to the backend optimization queue;

[0146] When it is not the first frame, obtaining map construction information according to the current frame information of the vehicle and joining the backend optimization queue;

[0147] When it is not the first frame, before matching the map construction information with the preset static map, it also includes:

[0148] Determine whether the static map is empty, and if so, initialize the static map according to the current frame information, and add the obtained static map to the backend optimization queue for matching with the map construction information.

[0149] In the fields of autonomous driving, robot navigation, etc., the back-end optimization queue refers to a queue system used to store and process data to be optimized in the data processing flow. In the technical solution provided in this embodiment, it is first determined whether the currently captured frame information is the first frame in the sequence. If it is the first frame, it is used as the initial reference frame and its information is added to the back-end optimization queue. The reference frame serves as the basis for subsequent frame matching and positioning. If the current frame is not the first frame, the map construction information will be obtained based on the information of this frame, that is, sensor data (such as lidar, camera, IMU) will be used to extract environmental features and build or update the map. Before matching the map construction information with the preset static map, check whether the static map is empty. If the static map is empty, the static map will be initialized using the current frame information and added to the back-end optimization queue to match it with the map construction information. Through the setting of the back-end optimization queue, complex data processing is allowed in the background, so as not to affect the real-time performance of the front end.

[0150] Therefore, on low-computing power platforms with limited computing resources and storage resources, this technical solution only adds the first frame as a reference frame to the back-end optimization queue when needed, which can avoid unnecessary data processing and save computing resources and storage space. For non-first frames, map construction information is directly obtained and added to the back-end optimization queue, which reduces repeated data processing steps, improves data processing efficiency, and thus reduces data processing time and energy consumption. When the static map is empty, the current frame information is used to dynamically initialize the static map, ensuring that even without pre-loading map data, it can quickly start and perform effective map construction and matching to achieve rapid response to subsequent vehicle positioning.

[0151] As an example, a vehicle positioning method based on a low computing power platform is provided for positioning a vehicle in a target area, comprising the following steps:

[0152] Calibrate the external parameters of the camera module, laser measurement module and inertial measurement module so that their mutual relations are fixed parameters;

[0153] Using the camera module to acquire image data of the vehicle, using the laser measurement module to acquire laser radar data of the vehicle, and using the inertial measurement module to acquire inertial measurement data of the vehicle; using the image data, the laser radar data, and the inertial measurement data as current frame information;

[0154] Determine whether the current frame information is the first frame. If it is the first frame, use the current frame information as the reference frame information and add it to the backend optimization queue;

[0155] When it is not the first frame, map construction information is obtained according to the current frame information of the vehicle and added to the back-end optimization queue; whether the static map is empty is determined, and if empty, the static map is initialized according to the current frame information, and the obtained static map is added to the back-end optimization queue for matching with the map construction information;

[0156] De-distorting the laser radar data according to the inertial measurement data, removing dynamic objects from the laser radar data after distortion processing, and obtaining map construction information; the current frame information includes image data, laser radar data and inertial measurement data;

[0157] Matching the map construction information with a preset static map to obtain a local map;

[0158] Based on the camera coordinate system, performing posture interpolation according to respective timestamps of the image data and the lidar data to determine a specific posture of the camera module in the local map;

[0159] Projecting the point cloud data in the local map onto a two-dimensional image plane corresponding to the image data according to the internal parameters of the camera module, and generating a camera photo containing depth information as depth image information;

[0160] Using a segmentation model of a low computing power platform to perform semantic segmentation of ground markings on the depth image information, and using the segmented point cloud with depth information as a ground marking point cloud;

[0161] Determine the current working mode of the vehicle, which includes a positioning mode and a mapping mode; when the working mode is the positioning mode, load the historical key point cloud to form a local historical map with only ground markings; match the ground marking point cloud with the local historical map, and output the repositioning posture of the vehicle; when the working mode is the mapping mode, associate the ground marking point cloud with the current position of the vehicle and save it as a ground marking key frame.

[0162] Among them, the vehicle includes at least one of an automatic inspection vehicle and an energy storage charging vehicle, the target area is an area where real-time dynamic measurement signals are missing, the segmentation model is a UNet network model, and the low computing power platform is built on the vehicle controller of the vehicle.

[0163] As another example, see Figure 3 , Figure 3 : is a flow chart of another vehicle positioning method provided in an embodiment of the present application. As a vehicle positioning method based on a low computing power platform, the method comprises the following steps:

[0164] 1.3D Lidar starts (starting the laser measurement module and reading Lidar data; starting the IMU program and reading IMU data), and determines whether it is the first frame;

[0165] 2. Yes, set it as the reference frame and add it to the backend optimization queue;

[0166] 3. No, dynamically build a reference frame and perform rough matching, then compare the differences, filter out dynamic objects, and then add them to the backend optimization queue; that is, fuse the Lidar and IMU data, remove distortion, and remove moving objects based on the reference frame, and add them to the backend optimization queue;

[0167] 4. Determine whether the static map is empty; if it is, use the current frame to initialize the static map and add it to the backend precision optimization queue for subsequent precision matching; if it is not, that is, if there is a static map, perform precision matching with the static map to obtain a more accurate Dof position (pose estimation result) and build a local map; the pose estimation result is added to the pose queue; where the pose queue is a data structure used to store pose data at a series of time points;

[0168] 5. The camera receives the photo (starts the camera module to obtain the photo as image data), and then interpolates the laser's historical posture to calculate its posture based on time;

[0169] 6. Determine whether it is successful; if unsuccessful, return / do not participate, during initialization, the pose queue is empty; if successful, project the point cloud to the camera according to the calibrated external parameters, and obtain a camera photo with partial depth information; that is, perform pose interpolation based on the received image information, according to the timestamp and the timestamp of the lidar, obtain the pose of the camera in the local image, and project the local point cloud onto the image according to the calibrated internal parameters to obtain an image with partial depth information;

[0170] 7. Segment ground markings based on UNet; mark the segmented ground marking point cloud as ground marking point cloud; it can be understood as semantic segmentation of the image, and for the segmented image, mark the part of the point cloud projection as semantic information;

[0171] 8. Determine whether it is in mapping mode; if it is in mapping mode, it is associated with the current position and saved as a ground marking keyframe; if it is in positioning mode, the historical key point cloud is loaded to form a local map with only ground markings, and the position is matched with the previous ground marking point cloud to output the repositioning pose.

[0172] Electronic device embodiments.

[0173] This embodiment provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor is configured to implement the steps of any one of the method embodiments when executing the computer program.

[0174] See also Figure 4 , Figure 4 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application.

[0175] The electronic device 10 may include, for example, at least one memory 11 , at least one processor 12 , and a bus 13 connecting different platform systems.

[0176] The memory 11 may include a (computer) readable medium in the form of a volatile memory, such as a random access memory (RAM) 111 and / or a cache memory 112 , and may further include a read-only memory (ROM) 113 .

[0177] The memory 11 also stores a computer program, which can be executed by the processor 12 so that the processor 12 implements the steps of any of the above methods.

[0178] The memory 11 may also include a utility 114 having at least one program module 115, such program module 115 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the implementation of a network environment.

[0179] Accordingly, the processor 12 may execute the above-mentioned computer program, and may execute the utility 114 .

[0180] The processor 12 may be implemented by one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic components.

[0181] Bus 13 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures.

[0182] The electronic device 10 may also communicate with one or more external devices such as a keyboard, a pointing device, a Bluetooth device, etc., and may also communicate with one or more devices that can interact with the electronic device 10, and / or communicate with any device (such as a router, a modem, etc.) that enables the electronic device 10 to communicate with one or more other computing devices. Such communication may be performed through an input / output interface 14. In addition, the electronic device 10 may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN) and / or a public network, such as the Internet) through a network adapter 15. The network adapter 15 may communicate with other modules of the electronic device 10 through the bus 13. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 10 in actual applications, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0183] System Embodiments.

[0184] This embodiment provides a vehicle positioning system, including the electronic device described in the electronic device embodiment, and also including an information acquisition device electrically connected to the electronic device; the information acquisition device includes a camera module, a laser measurement module and an inertial measurement module, the camera module is used to acquire image data, the laser measurement module is used to acquire laser radar data, and the inertial measurement module is used to acquire inertial measurement data. The electronic device is, for example, arranged on a vehicle controller of a vehicle as a part of the vehicle controller.

[0185] In one embodiment, it further includes a mode selection device, wherein the mode selection device is used to receive a user's instruction to switch the vehicle between a mapping mode and a positioning mode.

[0186] In the various implementation modes of the specification of the present application, the size of the serial number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the present application.

[0187] The terms "first", "second", "third", etc. (if any) in the specification and claims of the present application and the drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "corresponding to" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0188] This application is explained from the perspectives of purpose of use, effectiveness, progress and novelty, and has met the functional enhancement and usage requirements emphasized by the Patent Law. The above description and drawings of this application are only the preferred embodiments of this application, and are not intended to limit this application. Therefore, all structures, devices, features, etc. that are similar or identical to this application, that is, all equivalent replacements or modifications made in accordance with the scope of the patent application of this application, should fall within the scope of protection of the patent application of this application.

Claims

1. A vehicle positioning method based on a low computing power platform, characterized in that: The method for locating a vehicle in a target area includes the following steps: Acquiring map construction information according to current frame information of the vehicle; the current frame information includes image data, laser radar data and inertial measurement data; Matching the map construction information with a preset static map to obtain a local map; Based on a camera coordinate system, projecting the point cloud of the local map onto the image data, and obtaining a camera photo containing depth information as depth image information; Using a segmentation model of a low-computing-power platform to perform semantic segmentation of ground markings on the depth image information, using the segmented point cloud with depth information as a ground marking point cloud, and relocating the vehicle according to the ground marking point cloud; Wherein, the vehicle includes at least one of an automatic inspection vehicle and an energy storage charging vehicle, the target area is an area where real-time dynamic measurement signals are missing, and the segmentation model is a UNet network model; The method further includes: determining a current working mode of the vehicle, the current working mode including a positioning mode and a mapping mode; When the working mode is the positioning mode, repositioning the vehicle according to the ground marking point cloud includes: Loading the historical key point cloud to form a local historical map with only ground markings; matching the ground marking point cloud with the local historical map, and outputting the repositioned posture of the vehicle; When the working mode is the mapping mode, the ground marking point cloud is associated with the current position of the vehicle and saved as a ground marking key frame; The method further includes: determining whether the current frame information is the first frame, and if it is the first frame, taking the current frame information as reference frame information and adding it to a back-end optimization queue; the back-end optimization queue refers to a queue system used to store and process data to be optimized in a data processing flow; When it is not the first frame, obtaining map construction information according to the current frame information of the vehicle and joining the backend optimization queue; When it is not the first frame, before matching the map construction information with the preset static map, it also includes: Determine whether the static map is empty, and if so, initialize the static map according to the current frame information, and add the obtained static map to the backend optimization queue for matching with the map construction information.

2. The vehicle positioning method based on a low computing power platform according to claim 1 is characterized in that: The acquiring map construction information according to the current frame information of the vehicle includes: The laser radar data is subjected to dedistortion processing according to the inertial measurement data, and dynamic objects are removed from the laser radar data after the distortion processing to obtain map construction information.

3. The vehicle positioning method based on a low computing power platform according to claim 1, characterized in that: Before obtaining the map construction information, the method further includes: Calibrate the external parameters of the camera module, laser measurement module and inertial measurement module so that their mutual relations are fixed parameters; The camera module is used to acquire image data of the vehicle, the laser measurement module is used to acquire laser radar data of the vehicle, and the inertial measurement module is used to acquire inertial measurement data of the vehicle; and the image data, the laser radar data and the inertial measurement data are used as current frame information.

4. The vehicle positioning method based on a low computing power platform according to claim 3 is characterized in that: The process of projecting the point cloud in the local map to the image data comprises: Performing posture interpolation according to respective timestamps of the image data and the laser radar data to determine a specific posture of the camera module in the local map; According to the internal parameters of the camera module, the point cloud data in the local map is projected onto a two-dimensional image plane corresponding to the image data, and a camera photo containing depth information is generated as depth image information.

5. The vehicle positioning method based on a low computing power platform according to claim 4 is characterized in that: The point cloud of the local map only includes the laser point cloud of the ground, and projecting the point cloud data in the local map onto a two-dimensional image plane corresponding to the image data according to the internal parameters of the camera module includes: For each laser point cloud, first convert it according to the following formula: Then the transformed points are dedistorted to realize the projection of point cloud data. The dedistortion process corresponds to the following formula: Among them, K is the camera internal parameter matrix, the coefficient ( k 1, k 2, k 3) is the radial distortion, the coefficient ( p 1, p 2) is the tangential distortion, x and y are the coordinates of the laser point cloud, x corrected , y corrected Corresponding to the data after x and y distortion, r 2 Normalized distance from the laser point cloud to the image center; is the transformation matrix, which is the union of the rotation matrix and the translation vector. The rotation matrix is , the translation vector is ; is the point coordinate of the global point cloud at time k+1.

6. The vehicle positioning method based on a low computing power platform according to claim 1, characterized in that: The low computing power platform is built on a vehicle controller of the vehicle.

7. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to implement the steps of the method according to any one of claims 1 to 6 when executing the computer program.

8. A vehicle positioning system, characterized in that: The electronic device according to claim 7 also includes an information acquisition device electrically connected to the electronic device; the information acquisition device includes a camera module, a laser measurement module and an inertial measurement module, the camera module is used to acquire image data, the laser measurement module is used to acquire laser radar data, and the inertial measurement module is used to acquire inertial measurement data.

Images