Vehicle high-precision positioning system and method in large-scale real vehicle test

Abstract

The application relates to the technical field of vehicles, in particular to a high-precision positioning system and method for vehicles in large-scale real vehicle testing. The system comprises a plurality of network base stations, a plurality of test vehicles and an upper computer. The plurality of network base stations are arranged on a test road to form a local area network covering the test road. Each test vehicle broadcasts its position information in a map and vehicle attitude information through a protocol. The test vehicle collects video data around the vehicle through a plurality of cameras, shares a time signal to the local area network through a data sending unit, and detects acceleration data and angular velocity data of the vehicle through an inertial measurement unit. The upper computer is internally arranged with a bag-of-words model based on a pre-calibrated SLAM map construction to determine the position information of the vehicle in the map and the vehicle attitude information, and corrects the vehicle attitude information according to the acceleration data and the angular velocity data. Thus, the problems that related technologies have high costs in large-scale automatic driving testing and cannot fully verify required functions are solved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a high-precision vehicle positioning system and method for large-scale real vehicle testing. Background Technology

[0002] In real-time testing of autonomous vehicles, obtaining high-precision vehicle location information is crucial. Currently, the common approach involves installing high-precision positioning and measurement equipment on multiple test vehicles and uploading the data to a host computer via a local area network for real-time processing to obtain information such as relative position and speed between vehicles. However, this solution relies on expensive high-precision positioning equipment, resulting in high system costs and making it difficult to meet the needs of large-scale testing. Furthermore, in complex environments where satellite signals are easily blocked or interfered with, positioning accuracy and reliability significantly decrease. Therefore, there is an urgent need for a technical solution that does not rely on high-cost equipment and high-precision satellite positioning data, and can achieve high-precision, high-reliability positioning in complex environments. Summary of the Invention

[0003] This application provides a high-precision vehicle positioning system and method for large-scale real-vehicle testing to solve the problems of high cost and inability to fully verify the required functions in large-scale autonomous driving testing of related technologies.

[0004] The first aspect of this application provides a high-precision vehicle positioning system for large-scale real-vehicle testing, comprising the following steps: multiple network base stations, multiple test vehicles, and a host computer; wherein the multiple network base stations are set up on the test road, forming a local area network (LAN) that covers the test road; each test vehicle broadcasts its own location information and vehicle attitude information on a map via a protocol; the test vehicle includes multiple cameras, a GPS receiver, an inertial measurement unit (IMU), and a data transmission unit; wherein the multiple cameras are mounted on the test vehicle to collect video data around the vehicle, the video data including multiple frames of data; and the GPS receiver receives G... The PS time signal uses the GPS time signal as the timestamp of the current frame data. The data transmission unit shares the time signal with the local area network. The inertial measurement unit detects the vehicle's acceleration and angular velocity data. The host computer deploys a bag-of-words model based on a pre-calibrated SLAM map. The host computer obtains the current moment's video data from the local area network, extracts the image feature points of the current frame data from the current moment's video data, matches the map feature points from the bag-of-words model based on the implemented image feature points, determines the vehicle's position information and vehicle attitude information on the map based on the map feature points, and corrects the vehicle attitude information based on the acceleration and angular velocity data.

[0005] Optionally, multiple cameras are installed behind the test vehicle for debugging and calibration.

[0006] Optionally, the data transmission unit of the test vehicle packages its own location information on the map and vehicle attitude information into TCP / IP protocol data packets and sends the TCP / IP protocol data packets to the local area network.

[0007] Optionally, the host computer receives data from other test vehicles, calculates the relative distance, relative speed, and heading angle between the two vehicles based on the data and timestamps of the other test vehicles, generates a display image of the test vehicles based on the relative distance, relative speed, and heading angle, and displays the display image on the SLAM map.

[0008] Optionally, the host computer clusters the map feature points in the SLAM map and uses a bag-of-words model based on the clustering results, where the bag-of-words model assigns indices to the map feature points.

[0009] Optionally, the host computer calculates the pose transformation of the test vehicle using the least squares method, determines the relative position of the vehicle in the SLAM map based on the pose transformation, and transforms it to the world coordinate system according to the pre-calibrated transformation matrix to obtain the vehicle's position information and vehicle attitude information in the map.

[0010] Optionally, the SLAM map calibration process includes: constructing a SLAM map based on real vehicle test road data; extracting measured images from the real vehicle test road data; performing corner detection on the measured images; finding the corner positions of the calibration object in the measured images; and calculating the camera's intrinsic and extrinsic parameters and the transformation matrix between the map and the world coordinate space based on the actual size of the calibration object and its corner positions in the measured images, thereby completing the SLAM map calibration.

[0011] Optionally, a SLAM map is constructed based on real-vehicle test road data, including: collecting map information of the test route through a visual SLAM algorithm during real-vehicle testing, wherein the vehicle determines its own location information in the map while constructing the map during real-vehicle testing; and constructing a SLAM map based on the map information carrying the location information.

[0012] Optionally, a SLAM map is constructed based on real-vehicle test road data, including: scanning feature points on video image frames during real-vehicle testing; and constructing a SLAM map based on feature point matching between the current video image frame and adjacent video image frames.

[0013] The second aspect of this application provides a method for high-precision vehicle positioning in large-scale real-vehicle testing. The method is applied to test vehicles in the high-precision vehicle positioning system of the first aspect of large-scale real-vehicle testing. Each test vehicle broadcasts its own position information and vehicle attitude information on the map via a protocol. The test vehicle includes multiple cameras, a GPS receiver, an inertial measurement unit, a data transmission unit, and a host computer. The multiple cameras are mounted on the test vehicle to collect video data around the vehicle. The video data includes multiple frames. The GPS receiver receives GPS time signals and uses the GPS time signals as the timestamp of the current frame data. The data transmission unit shares the time signals to a local area network. The inertial measurement unit detects the vehicle's acceleration and angular velocity data. The host computer is configured to perform the following steps: constructing a bag-of-words model based on a pre-calibrated SLAM map; extracting image feature points from the current frame data in the current video data; matching map feature points from the bag-of-words model based on the implemented image feature points; determining the vehicle's position information and vehicle attitude information on the map based on the map feature points; and correcting the vehicle attitude information based on the acceleration and angular velocity data.

[0014] Therefore, this application has at least the following beneficial effects:

[0015] This application utilizes a local area network (LAN) formed by multiple network base stations deployed on the test road to cover the test road. Each test vehicle broadcasts its position and attitude information on the map via a protocol. The test vehicle collects video data from its surroundings via multiple cameras, and the time signal is shared to the LAN through a data transmission unit. An inertial measurement unit detects the vehicle's acceleration and angular velocity data. A bag-of-words model based on a pre-calibrated SLAM map is deployed in the host computer to determine the vehicle's position and attitude information on the map. The vehicle's attitude information is corrected based on the acceleration and angular velocity data. This method does not rely on expensive high-precision positioning equipment; it calculates the vehicle's high-precision position by acquiring video data, reducing testing costs. It can be used in large-scale test vehicles, providing sufficient verification for the performance evaluation of autonomous driving test systems. Therefore, it solves the problems of high cost and inability to fully verify required functions in large-scale autonomous driving testing of related technologies.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0018] Figure 1This is a block diagram of a high-precision vehicle positioning system for large-scale real-vehicle testing according to an embodiment of this application;

[0019] Figure 2 This is a flowchart illustrating a high-precision vehicle positioning method for large-scale real-vehicle testing according to an embodiment of this application.

[0020] Figure 3 This is a flowchart illustrating a high-precision vehicle positioning method for large-scale real-vehicle testing provided according to an embodiment of this application. Detailed Implementation

[0021] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0022] The following description, with reference to the accompanying drawings, illustrates a high-precision vehicle positioning system and method for large-scale real-vehicle testing according to embodiments of this application. Addressing the issues of high cost and insufficient verification of required functions in large-scale autonomous driving testing of related technologies mentioned in the background, this application provides a high-precision vehicle positioning system for large-scale real-vehicle testing. In this system, a local area network (LAN) is formed by multiple network base stations set up on the test road to cover the test road. Each test vehicle broadcasts its position information and vehicle attitude information on the map via a protocol. The test vehicle collects video data from its surroundings through multiple cameras, and the time signal is shared to the LAN through a data transmission unit. An inertial measurement unit detects the vehicle's acceleration and angular velocity data. A bag-of-words model based on a pre-calibrated SLAM map is deployed in the host computer to determine the vehicle's position and attitude information on the map. The vehicle attitude information is corrected based on the acceleration and angular velocity data. This system does not rely on expensive high-precision positioning equipment; instead, it calculates the vehicle's high-precision position by acquiring video data, reducing testing costs. It can be used in large-scale test vehicle testing, providing sufficient verification for the performance evaluation of autonomous driving testing systems. Therefore, it solves the problems of high cost and insufficient verification of required functions in large-scale autonomous driving testing of related technologies.

[0023] Specifically, Figure 1 This is a block diagram illustrating a high-precision vehicle positioning system for large-scale real-vehicle testing, as provided in an embodiment of this application. Figure 1As shown, the high-precision vehicle positioning system 10 in this large-scale real vehicle test includes: multiple network base stations 101, multiple test vehicles 102 and host computer 103. Among them, the test vehicle 102 includes multiple cameras 1021, GPS (Global Positioning System) receiver 1022, inertial measurement unit 1023 and data transmission unit 1024.

[0024] Multiple network base stations 101 are installed on the test road, forming a local area network (LAN) that covers the test road. Each test vehicle 102 broadcasts its location and attitude information on the map via a protocol. The test vehicle 102 includes multiple cameras 1021, a GPS receiver 1022, an inertial measurement unit 1023, and a data transmission unit 1024. The multiple cameras 1021 are mounted on the test vehicle 102 to collect video data around the vehicle, including multiple frames. The GPS receiver 1022 receives GPS time signals and uses them as timestamps for the current frame. The data transmission unit 1024 shares the time signals with the LAN. The inertial measurement unit 1023 detects the vehicle's acceleration and angular velocity data. The host computer 103 is equipped with a pre-calibrated SLAM (Simultaneous Localization and Interference) system. Mapping (real-time localization and map building) uses a bag-of-words model for map building. The host computer 103 obtains the current video data from the local area network, extracts the image feature points of the current frame data in the current video data, matches map feature points from the bag-of-words model based on the implemented image feature points, determines the vehicle's position information and vehicle attitude information on the map based on the map feature points, and corrects the vehicle attitude information based on acceleration data and angular velocity data.

[0025] Among them, network base station 101 refers to a wireless network access point fixedly set up on the test road. Multiple base stations work together to form a local area network (LAN) covering the entire test area, ensuring that the test vehicle 102 can continuously and stably access the network during operation. The LAN is a private communication network (such as using DLNA (Direct Local Area Network) technology) constructed by multiple network base stations 101 on the test road. It is used to transmit data such as the vehicle's position, attitude, and timestamps between the test vehicle 102 and the host computer 103, achieving low-latency and high-reliability data exchange. The protocol refers to the rules and formats followed by the data communication between the vehicle and the host computer 103 (such as TCP / IP (Transmission Control Protocol / Internet Protocol)). The Internet Protocol (IP) ensures that vehicles can broadcast their position and attitude information to the local area network according to a unified standard; the GPS receiver 1022 is a device installed on the vehicle to receive satellite signals; the SLAM map refers to a high-precision map containing rich environmental features generated by scanning and mapping the test road environment in advance using a camera through visual SLAM technology; the bag-of-words model refers to an algorithm model for image retrieval and matching, which can cluster and index all feature points in the pre-built SLAM map to form a vocabulary. When the test vehicle 102 acquires a new image, its extracted feature points can be quickly compared with this "vocabulary" to achieve efficient and accurate feature matching; to extract image feature points from the current frame data in the video data at the current moment, methods such as ORB (Oriented FAST and Rotated BRIEF, where FAST (Features from Accelerated Segment Test) is an algorithm for detecting feature points in an image, and BRIEF (Binary Robust Independent Element) can be used. Features (binary robust independent basic features) are algorithms used to describe detected feature points. Algorithms such as these extract significant and stable visual feature points from images.

[0026] It is understood that in this embodiment of the application, multiple network base stations 101 are deployed along the test road to construct a local area network covering the entire road. Each test vehicle is equipped with multiple cameras 1021, a GPS receiver 1022, an inertial measurement unit 1023, and a data transmission unit 1024, which are connected to a host computer 103. The host computer 103 pre-stores a high-precision map constructed using SLAM technology and calibrated with world coordinates, and establishes a bag-of-words model based on the feature points of this map. Video data, including multiple frames, is continuously collected from the area around the vehicle by the multiple cameras 1021, and is compared with the bag-of-words model in the host computer 103. The vehicle's precise position and initial attitude in the SLAM map can be calculated based on the matching results. Combined with the acceleration and angular velocity data measured in real time by the inertial measurement unit 1023, the vehicle's attitude information is dynamically corrected and optimized to further improve the continuity and reliability of positioning. At the same time, the GPS receiver 1022 provides accurate GPS time as the timestamp for each frame of data to ensure time synchronization of data from multiple vehicles. Finally, the final position and attitude information calculated by the vehicle is broadcast to the host computer 103 via the local area network by the data transmission unit 1024 according to a specific communication protocol, so that the host computer can perform comprehensive analysis and display of the relative status of multiple vehicles.

[0027] In this embodiment, a multi-channel camera 1021 is installed behind the test vehicle 102 for debugging and calibration.

[0028] Among these steps, when installing multiple cameras 1021, it is important to ensure that the camera installation positions match those of the cameras on the data collection vehicle previously used to build the SLAM map, so as to obtain images from the same perspective. Debugging refers to physical and parameter adjustments to the camera hardware, such as checking whether the camera is securely installed and whether the exposure parameters are appropriate, to ensure that the camera can work normally and stably. Calibration is a more precise technical process aimed at ensuring that the cameras on the test vehicle 102 have the same installation positions and perspectives as the cameras on the data collection vehicle previously used to build the SLAM map, in order to guarantee the accuracy of subsequent image feature extraction and map matching.

[0029] It is understood that after installing the multi-camera 1021 onto the test vehicle 102 in this embodiment of the application, it is necessary to debug and calibrate it to ensure that all cameras are physically securely installed, have unobstructed views, and are in normal working condition. Furthermore, the internal and external parameters of each camera must be accurately determined, and it is especially important to ensure that its installation position and viewing angle are completely consistent with the cameras on the data acquisition vehicle previously used to build the SLAM map. This is to ensure that the images acquired in real time by the test vehicle 102 have the exact same viewing angle as the images on which the pre-built SLAM map is based, thereby avoiding errors caused by viewing angle differences in the subsequent positioning and matching process, and laying the foundation for achieving high-precision position matching.

[0030] In this embodiment of the application, the data transmission unit 1024 of the test vehicle 102 packages its own location information in the map and vehicle attitude information into TCP / IP protocol data packets, and sends the TCP / IP protocol data packets to the local area network.

[0031] Among them, location information refers to the precise coordinates of the vehicle in the map coordinate system calculated by matching the image features captured by the vehicle's camera with the pre-built and calibrated SLAM map; vehicle attitude information refers to the direction and angle state of the vehicle in three-dimensional space, usually including yaw angle, pitch angle and roll angle, describing the vehicle's orientation and tilt degree; TCP / IP protocol data packets refer to data units formatted and encapsulated according to the TCP / IP network communication standard.

[0032] It is understood that after the test vehicle 102 in this embodiment of the application completes the matching with the pre-built SLAM map using camera images and calculates its precise position and vehicle attitude in the map, the data transmission unit 1024 on the test vehicle 102 will process this key information, encapsulate the position information and vehicle attitude information in the standard format of the TCP / IP network communication protocol to form a data packet, and send the data packet out through the local area network connected to the test vehicle 102, so as to broadcast the real-time status information of the test vehicle 102 to the host computer 103 or other related devices in the network for subsequent centralized monitoring, analysis and multi-vehicle collaborative calculation.

[0033] In this embodiment, the host computer 103 receives data from other test vehicles 102, calculates the relative distance, relative speed, and heading angle between the two vehicles based on the data and timestamps of the other test vehicles 102, generates a display image of the test vehicle 102 based on the relative distance, relative speed, and heading angle, and displays the display image on the SLAM map.

[0034] Among them, relative distance refers to the straight-line distance between the current position coordinates of the two test vehicles 102; relative speed refers to the speed of one test vehicle 102 relative to another test vehicle 102, including speed magnitude and direction, which can be calculated by the vector difference of the absolute speeds of the two vehicles; heading angle refers to the angle between the vehicle's forward direction and geographic true north (or a fixed direction in the map coordinate system), used to describe the vehicle's heading; display image refers to presenting the calculated relative status information (such as relative distance, speed, and heading) in a graphical way so as to intuitively monitor the dynamic relationship between the test vehicles 102.

[0035] It is understood that the host computer 103 in this embodiment of the application acts as a data processing center, receiving the position and attitude information and their corresponding timestamps broadcast from other test vehicles 102 in the local area network. By parsing these data and synchronizing with precise timestamps, the host computer 103 can calculate key dynamic parameters such as the relative distance, relative speed and heading angle between the test vehicle 102 and other test vehicles 102 in real time. Based on these calculation results, it generates intuitive graphical display information and overlays these display graphics on a pre-built SLAM high-precision map background, thereby providing testers with a clear and real-time view of the multi-vehicle collaborative operation situation.

[0036] In this embodiment, the host computer 103 clusters the map feature points in the SLAM map and uses a bag-of-words model based on the clustering results, wherein the bag-of-words model assigns an index to the map feature points.

[0037] Clustering is a data mining algorithm that groups a large number of similar feature points according to their similarity. In the bag-of-words model, all image feature points are input into a clustering algorithm (such as K-means, a classic unsupervised machine learning clustering algorithm), and similar map feature points are classified into the same class.

[0038] It is understood that the host computer 103 in this embodiment of the application processes the map feature points in the pre-built SLAM map. Specifically, it uses a clustering algorithm to group the descriptors of these feature points according to their similarity, forming multiple cluster centers. Each cluster center represents a unique class. Based on these clustering results, the host computer 103 constructs a bag-of-words model and assigns a unique numerical index to each class in the model, thereby forming an efficient visual vocabulary, realizing the construction of the bag-of-words model, and storing the constructed bag-of-words model in the host computer 103 for subsequent testing. This allows for rapid matching of feature points in the real-time acquired images of the test vehicle 102 with feature points in the SLAM map, achieving high-precision vehicle positioning.

[0039] In this embodiment, the host computer 103 calculates the pose transformation of the test vehicle 102 using the least squares method, determines the relative position of the vehicle in the SLAM map based on the pose transformation, and transforms it to the world coordinate system according to the pre-calibrated transformation matrix to obtain the vehicle's position information and vehicle attitude information in the map.

[0040] Least squares is a mathematical optimization technique used to find the best function match for a set of data. Its core idea is to minimize the sum of squares of the errors between the observed data and the model's predicted data. In localization, by finding an optimal pose transformation, the total projection error between the feature points extracted from the current image and the feature points matched in the SLAM map is minimized, thus obtaining the most accurate vehicle pose estimation. Pose transformation refers to the position and attitude changes of the test vehicle 102 from one coordinate system to another, including translation and rotation. The pre-calibration transformation matrix refers to the mathematical matrix used to transform between the SLAM map coordinate system and the real-world coordinate system, calculated during the system preparation phase through the map calibration process (such as using the Zhang Zhengyou calibration method (a classic camera calibration method based on a planar checkerboard, which is one of the most widely used and practical camera calibration techniques in the field of computer vision) in conjunction with calibration objects such as checkerboards).

[0041] It is understood that, in this embodiment of the application, after the camera of the test vehicle 102 acquires the current image and extracts feature points, the host computer 103 will match these feature points with feature points in the pre-built SLAM map to find a set of corresponding matching point pairs. In order to accurately calculate the current pose of the vehicle, the host computer 103 will use the least squares method as an optimization algorithm to solve an optimal pose transformation, so that the geometric error between the matching point pairs is minimized, and directly give the relative position and initial pose of the test vehicle 102 in the SLAM map coordinate system. In order to obtain a positioning result with actual geographical significance, it is also necessary to transform this relative pose to the global world coordinate system. This transformation is accomplished by applying the pre-calibrated transformation matrix that has been determined in the system calibration stage. Through the above steps, the accurate position information and complete vehicle pose information of the vehicle in the real geographical environment are obtained.

[0042] In this embodiment of the application, the SLAM map calibration process includes: constructing a SLAM map based on real vehicle test road data; extracting the measured images from the real vehicle test road data; performing corner detection on the measured images; finding the corner positions of the calibration object in the measured images; and calculating the intrinsic and extrinsic parameters of the camera and the transformation matrix between the map and the world coordinate space based on the actual size of the calibration object and its corner position in the measured images, thereby completing the SLAM map calibration.

[0043] Among them, real-vehicle test road data refers to the original video or image sequences recorded by a data collection vehicle equipped with cameras and other sensors while driving on actual test roads, which is the original material for building SLAM maps; calibration objects refer to objects placed on test roads with known precise geometric shapes and sizes, whose corner points have known three-dimensional coordinates in the physical world, serving as a bridge connecting image pixel coordinates and real-world coordinates; corner detection refers to the process of automatically identifying and locating the pixel coordinates of the corner points of calibration objects in the test images using computer vision algorithms.

[0044] It is understood that the SLAM map calibration process in this application embodiment first requires using image data collected on the actual vehicle test road to construct an initial SLAM map containing environmental features through a visual SLAM algorithm. In order to associate this map with the real geographic coordinates, calibration is required. Specifically, calibration objects of known size are placed on the test road, and measured images containing these calibration objects are extracted from the actual vehicle test road data used to construct the map. Corner detection is performed on these images to accurately find the pixel positions of the corner points of the calibration objects in the images. Using these known physical dimensions, the detected pixel positions, and algorithms such as Zhang Zhengyou calibration method, the intrinsic and extrinsic parameters of the camera are calculated simultaneously to establish a transformation matrix between the SLAM map coordinate system and the real world coordinate system, thereby realizing the calibration of the SLAM map. This gives the map geographic reference capability, which can be used for the subsequent location positioning of the test vehicle 102.

[0045] In this embodiment of the application, constructing a SLAM map based on real vehicle test road data includes: collecting map information of the test route through a visual SLAM algorithm during real vehicle testing, wherein the real vehicle determines its own location information in the map while constructing the map; and constructing a SLAM map based on the map information carrying the location information.

[0046] It is understood that the process of constructing a SLAM map in this application embodiment begins with data collection on a preset real vehicle test route. During this process, the positioning information of the collecting vehicle relative to the map being formed is calculated in real time, namely the vehicle's movement trajectory and attitude. The map information generated during the collection process, which contains rich environmental features, is integrated with its corresponding positioning information that records the vehicle's exploration path to form a SLAM map.

[0047] In this embodiment of the application, constructing a SLAM map based on real vehicle test road data includes: scanning feature points on video image frames during real vehicle testing; and constructing a SLAM map based on feature point matching between the current video image frame and adjacent video image frames.

[0048] It is understood that the process of constructing the SLAM map in this application embodiment is accomplished during real vehicle testing, using video data recorded by the acquisition vehicle while it is driving on the test road. Specifically, feature points are scanned for each frame of the video stream, and algorithms such as ORB are used to extract significant and stable visual feature points in the image. Through feature point matching technology, feature points in the current video image frame are associated and paired with feature points in adjacent video image frames. By analyzing the positional changes of these matched point pairs between consecutive frames, the SLAM algorithm can simultaneously calculate the vehicle's own trajectory and reconstruct the structure of the environment, thereby incrementally constructing a SLAM map containing environmental features.

[0049] The high-precision vehicle positioning system for large-scale real-vehicle testing proposed in this application uses a local area network (LAN) composed of multiple network base stations set up on the test road to cover the test road. Each test vehicle broadcasts its own position information and vehicle attitude information on the map through a protocol. The test vehicle collects video data around the vehicle through multiple cameras and shares the time signal to the LAN through a data transmission unit. The inertial measurement unit detects the vehicle's acceleration and angular velocity data. The host computer deploys a bag-of-words model based on a pre-calibrated SLAM map to determine the vehicle's position information and vehicle attitude information on the map. The vehicle attitude information is corrected based on the acceleration and angular velocity data. It does not rely on expensive high-precision positioning equipment and completes the calculation of the vehicle's high-precision position by acquiring video, which reduces the testing cost. It can be used in the testing of large-scale test vehicles and provides sufficient verification for the performance evaluation of autonomous driving test systems.

[0050] The following is a detailed description of a high-precision vehicle positioning system in large-scale real-vehicle testing, using a specific example.

[0051] This embodiment includes the following parts:

[0052] 1. Equipment installation and data acquisition:

[0053] Camera Installation: Install multiple cameras at the front, rear, left, and right of the vehicle to ensure coverage of key areas around the vehicle. Adjust and calibrate the cameras to ensure the clarity and accuracy of the captured images.

[0054] Image acquisition: During vehicle operation, the camera acquires images in real time at a set frame rate (e.g., 30fps).

[0055] Image preprocessing: The acquired images are subjected to grayscale conversion, noise reduction (such as Gaussian filtering), and enhancement (such as histogram equalization) to improve image quality.

[0056] Feature extraction: Use the selected feature extraction algorithm (such as ORB algorithm) to extract feature points from the preprocessed image and record the location and descriptor of the feature points.

[0057] 2. Time synchronization

[0058] GPS Receiver Installation: Install a high-precision GPS receiver on the vehicle to ensure stable GPS signal reception. The GPS time synchronization module reads the time signal received by the GPS receiver as the timestamp of the current frame for subsequent data processing.

[0059] The calibrated time information is transmitted to other vehicles via the local area network. At the same time, a timestamp is added to each data frame during data acquisition and transmission to ensure the time order and consistency of the data.

[0060] 3. Video frame features and matching:

[0061] Bag-of-words model construction: Cluster the feature points in the pre-built SLAM map to construct a bag-of-words model, and assign an index to each feature point.

[0062] Real-time matching: The feature points of the images captured by the camera are matched with the bag-of-words model to find matching feature point pairs.

[0063] Pose calculation: Using matched feature point pairs, the rotation and translation transformations of the vehicle are calculated through optimization algorithms such as the least squares method to obtain the vehicle's relative position and pose in the map.

[0064] 3. SLAM Mapping Implementation

[0065] Calibration object placement: Place multiple checkerboard-patterned calibration objects at different locations along the test route to ensure that the calibration objects can be clearly captured by the camera under different angles and lighting conditions.

[0066] Image acquisition: Images of the calibration object are acquired using a camera installed on the vehicle, taking multiple sets of images from different angles and positions.

[0067] 4. Inertial Measurement Unit Information Acquisition

[0068] Inertial Measurement Unit Installation: Install the inertial measurement unit (IMU) near the vehicle's center of gravity to ensure accurate measurement of the vehicle's motion. Initialize and calibrate the IMU to acquire initial acceleration and angular velocity data.

[0069] Data acquisition: The inertial measurement unit acquires the vehicle's acceleration and angular velocity data in real time.

[0070] Data Processing: Using selected Kalman filtering algorithms, the acquired inertial measurement unit (IMU) data is processed based on the vehicle's motion model and sensor measurement model. In the prediction step, the current state is predicted based on the state at the previous moment. In the update step, the prediction results are corrected using newly acquired sensor data to obtain a more accurate vehicle state estimate.

[0071] Corner detection: Using algorithms such as Zhang Zhengyou's calibration method, corner detection is performed on the acquired images to find the corner positions of the calibration objects in the images.

[0072] Parameter calculation: Based on the actual size of the calibration object and its corner position in the image, calculate the camera's intrinsic and extrinsic parameters as well as the transformation matrix between the map and the world coordinate space to complete the map calibration.

[0073] 5. Implementation of Local Area Network Transmission

[0074] A large-scale local area network is implemented using a fixed-area mesh network on the test road, covering the entire test road. The host computer of each vehicle is configured to enable seamless or low-latency switching between multiple network base stations. Each vehicle broadcasts its location information and vehicle attitude information on the map through a specific protocol, which are then processed and calculated by the host computer.

[0075] Specifically, such as Figure 2 As shown, the specific process of this embodiment is as follows:

[0076] 1. SLAM Map Building

[0077] By acquiring real-world test road data (including video image frames, etc.), the visual SLAM algorithm is used to scan and match feature points in the video image frames, gradually constructing a SLAM map containing environmental features, and simultaneously determining the vehicle's location information in the map, thus obtaining a SLAM map with location information.

[0078] 2. SLAM Mapping

[0079] Based on the corner detection results in the measured image and the actual size of the calibration object, the intrinsic and extrinsic parameters of the camera and the transformation matrix between the map and the world coordinate space are calculated to complete the calibration of the SLAM map.

[0080] 3. Test vehicle deployment map

[0081] The calibrated SLAM map was deployed onto the test vehicle as the basis for subsequent localization and navigation.

[0082] 4. Test vehicle video reading

[0083] The test vehicle reads video image frames in real time.

[0084] 5. Feature Extraction Module

[0085] The system receives video image frames read in real time from the test vehicle, extracts feature points from each frame, and generates a set of feature points describing the current environment.

[0086] 6. Feature Matching Module

[0087] The system receives the feature point set of the current frame and the SLAM map deployed on the test vehicle. It then matches the feature points of the current frame with the feature points in the SLAM map to determine the position of the test vehicle on the map.

[0088] 7. Test vehicle map location

[0089] The test vehicle obtains its map location.

[0090] 8. Calculation of coordinate transformation matrix

[0091] Based on the test vehicle's location information in the SLAM map and the transformation matrix between the map and world coordinate space, the test vehicle's position in the real geographic coordinate system is calculated.

[0092] 9. High-precision positioning output

[0093] Output the high-precision location information of the test vehicle in the real geographic coordinate system for use by subsequent modules.

[0094] 10. Local Area Network Transmission

[0095] High-precision location information is transmitted to the central host computer via a local area network.

[0096] 11. Overall host computer calculation

[0097] The host computer uses the high-precision location information received from the local area network to calculate the global location information.

[0098] 12. Global location information

[0099] The global position information calculated by the host computer is used for subsequent applications such as navigation and path planning.

[0100] In summary, the entire process begins with the construction of the SLAM map, and goes through multiple steps such as calibration, deployment, real-time feature extraction and matching, coordinate transformation, high-precision location output, local area network transmission, and overall host computer calculation, ultimately completing the global test vehicle location calculation.

[0101] Next, referring to the accompanying drawings, a method for high-precision vehicle positioning in large-scale real-vehicle testing according to an embodiment of this application is described.

[0102] Figure 3This is a block diagram illustrating a high-precision vehicle positioning method for large-scale real-vehicle testing according to an embodiment of this application. The method is applied to test vehicles in the aforementioned high-precision vehicle positioning system for large-scale real-vehicle testing. Each test vehicle broadcasts its own location information and vehicle attitude information on the map via a protocol. The test vehicle includes multiple cameras, a GPS receiver, an inertial measurement unit, a data transmission unit, and a host computer. The multiple cameras are mounted on the test vehicle to collect video data around the vehicle. The video data includes multiple frames. The GPS receiver receives GPS time signals and uses these signals as the timestamps of the current frame. The data transmission unit shares the time signals to the local area network. The inertial measurement unit detects the vehicle's acceleration and angular velocity data, such as... Figure 3 As shown, the host computer is configured to perform the following steps:

[0103] In step S201, a bag-of-words model is constructed based on the pre-calibrated SLAM map.

[0104] It is understood that the embodiments of this application construct a bag-of-words model based on a pre-calibrated SLAM map. This model extracts and clusters visual feature points in the map to form words, so that in the subsequent localization process, the image features captured by the vehicle camera can be quickly matched with the features in the map, thereby achieving efficient and accurate location recognition.

[0105] In step S202, image feature points of the current frame data in the video data at the current moment are extracted.

[0106] It is understood that, in the embodiments of this application, at the current moment, visual feature points of the latest frame image are extracted from the video stream captured by the test vehicle camera using a specific algorithm (such as ORB), and used for subsequent matching with a pre-built SLAM map to determine the precise location of the vehicle.

[0107] In step S203, map feature points are matched from the bag-of-words model based on the implemented image feature points.

[0108] It is understood that the embodiments of this application utilize the feature points extracted from the current image frame to perform fast retrieval and matching in a pre-built bag-of-words model, find the class with the most similarity, and thus determine the map feature points corresponding to these image feature points in the labeled SLAM map, thereby realizing the association between the vehicle's current position and the map.

[0109] In step S204, the vehicle's position information and attitude information on the map are determined based on map feature points, and the vehicle attitude information is corrected based on acceleration data and angular velocity data.

[0110] It is understood that, in this embodiment of the application, the correspondence between the map feature points obtained by matching and the current image feature points is used to calculate the vehicle's pose transformation relative to the SLAM map using optimization algorithms, such as the least squares method, thereby determining the vehicle's precise position and initial attitude information in the map. Subsequently, the calculated vehicle attitude is dynamically corrected and filtered by combining the vehicle acceleration and angular velocity data collected in real time by the inertial measurement unit, so as to improve the continuity and accuracy of attitude estimation, and enhance the robustness of the system, especially when the image information is temporarily unstable.

[0111] It should be noted that the foregoing explanation of the vehicle high-precision positioning system embodiment in large-scale real vehicle testing also applies to the vehicle high-precision positioning method in large-scale real vehicle testing of this embodiment, and will not be repeated here.

[0112] The high-precision vehicle positioning method for large-scale real-vehicle testing proposed in this application involves using multiple network base stations set up on the test road to form a local area network covering the test road. Each test vehicle broadcasts its own position information and vehicle attitude information on the map via a protocol. The test vehicle collects video data around the vehicle through multiple cameras, and shares the time signal to the local area network through a data transmission unit. An inertial measurement unit detects the vehicle's acceleration and angular velocity data. A bag-of-words model based on a pre-calibrated SLAM map is deployed in the host computer to determine the vehicle's position information and vehicle attitude information on the map. The vehicle attitude information is corrected based on the acceleration and angular velocity data. This method does not rely on expensive high-precision positioning equipment and completes the calculation of the vehicle's high-precision position by acquiring video, reducing testing costs. It can be used in the testing of large-scale test vehicles and provides sufficient verification for the performance evaluation of autonomous driving test systems.

[0113] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0114] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0115] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0116] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0117] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A high-precision vehicle positioning system for large-scale real-vehicle testing, characterized in that, include: Multiple network base stations, multiple test vehicles, and host computers, among which, The multiple network base stations are set up on the test road, and the multiple network base stations form a local area network. The local area network covers the test road, and each test vehicle broadcasts its own location information and vehicle attitude information on the map through a protocol. The test vehicle includes multiple cameras, a GPS receiver, an inertial measurement unit, and a data transmission unit. The multiple cameras are mounted on the test vehicle to collect video data around the vehicle. The video data includes multiple frames. The GPS receiver receives GPS time signals and uses the GPS time signals as the timestamp of the current frame data. The data transmission unit shares the time signals to the local area network. The inertial measurement unit detects the vehicle's acceleration and angular velocity data. The host computer is equipped with a bag-of-words model built based on a pre-calibrated SLAM map. The host computer obtains the video data at the current moment from the local area network, extracts the image feature points of the current frame data in the video data at the current moment, matches map feature points from the bag-of-words model based on the image feature points, determines the vehicle's position information and vehicle attitude information on the map based on the map feature points, and corrects the vehicle attitude information based on the acceleration data and the angular velocity data.

2. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 1, characterized in that, The multi-channel camera is installed behind the test vehicle, and the multi-channel camera is debugged and calibrated.

3. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 1, characterized in that, The data transmission unit of the test vehicle packages its own location information on the map and vehicle attitude information into TCP / IP protocol data packets, and sends the TCP / IP protocol data packets to the local area network.

4. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 1, characterized in that, The host computer receives data from other test vehicles, calculates the relative distance, relative speed, and heading angle between the two vehicles based on the data from other test vehicles and the timestamp, generates a display image of the test vehicles based on the relative distance, relative speed, and heading angle, and displays the display image on the SLAM map.

5. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 1, characterized in that, The host computer clusters the map feature points in the SLAM map and constructs the bag-of-words model based on the clustering results, wherein the bag-of-words model assigns indexes to the map feature points.

6. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 1, characterized in that, The host computer calculates the pose transformation of the test vehicle using the least squares method, determines the relative position of the vehicle in the SLAM map based on the pose transformation, and transforms it to the world coordinate system according to the pre-calibrated transformation matrix to obtain the vehicle's position information and vehicle attitude information in the map.

7. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 1, characterized in that, The calibration process of the SLAM map includes: SLAM maps are constructed based on real-world test road data. Extract the measured images from the real vehicle test road data, perform corner detection on the measured images, find the corner positions of the calibration object in the measured images, and calculate the camera's intrinsic and extrinsic parameters and the transformation matrix between the map and the world coordinate space based on the actual size of the calibration object and the corner positions in the measured images to complete the calibration of the SLAM map.

8. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 7, characterized in that, The construction of the SLAM map based on real-vehicle test road data includes: During the real vehicle test, the visual SLAM algorithm is used to collect map information of the test route. The real vehicle test constructs the map and determines its own location information in the map. The SLAM map is constructed based on map information carrying the location information.

9. The high-precision vehicle positioning system for large-scale real-vehicle testing according to claim 7, characterized in that, The construction of the SLAM map based on real-vehicle test road data includes: Feature point scanning was performed on video image frames during real vehicle testing; A SLAM map is constructed by matching feature points of the current video image frame with those of adjacent video image frames.

10. A method for high-precision vehicle positioning in large-scale real-vehicle testing, characterized in that, The method is applied to the test vehicle in the high-precision vehicle positioning system of the large-scale real vehicle test as described in any one of claims 1-9.

Images