A kind of well mine unmanned vehicle speed control method based on vehicle lane cloud cooperation

Abstract

The application discloses a kind of well mine unmanned vehicle speed control methods based on vehicle lane cloud cooperation, it is related to vehicle control: acquisition vehicle end sensor data;Vehicle end obstacle data set is generated;Collect roadside environment perception data, and generate roadside obstacle data set according to roadside environment perception data through perception fusion algorithm;Global obstacle map is constructed, and the obstacle data in the preset range is extracted with each vehicle position as center, and the vehicle obstacle map of each vehicle is generated;Based on global obstacle map and all vehicle state data, the global expected speed of each vehicle is calculated;According to vehicle end obstacle data set, roadside obstacle data set and the vehicle obstacle map issued by cloud, vehicle end expected speed is calculated;According to vehicle end expected speed, map speed limit and global expected speed, the final control speed is generated.The application improves traffic efficiency through the three-layer speed control architecture of cloud global expected speed-vehicle end expected speed-final control speed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicle control, in particular to a mine unmanned vehicle speed control method based on vehicle-lane-cloud collaboration. BACKGROUND

[0002] With the advancement of intelligent mine construction, mine unmanned technology has become a key technology to improve mine safety and production efficiency. The mine roadway environment has the characteristics of narrow space, frequent turning, poor lighting conditions, and frequent multi-vehicle intersection, which puts high requirements on the speed control of unmanned vehicles. As a core function of unmanned driving, speed control directly affects the safety, traffic efficiency and transportation efficiency of vehicles. However, the special closed environment and complex working conditions of the mine make it difficult to directly apply the speed control method of ground unmanned vehicles, and it is urgent to develop special speed control technology suitable for underground environment.

[0003] The existing speed control technology of mine unmanned vehicles mainly has the following problems:

[0004] (1) Single speed decision source, lack of multi-level collaboration mechanism: existing technology mainly relies on vehicle-mounted sensors for local perception and speed decision, lacking global perspective for speed optimization; the vehicle only controls the speed based on the obstacle information within its own perception range, and cannot know the traffic conditions in front of the roadway in advance; there is a lack of collaboration mechanism between cloud global optimization and real-time control at the vehicle end, resulting in unreasonable speed control in complex scenes such as multi-vehicle intersection.

[0005] (2) Incomplete speed constraint mechanism, insufficient safety protection: existing methods usually only consider a single constraint condition (such as only based on obstacle distance or only based on preset speed curve); lack of multi-constraint mechanism integrating real-time perception, static map speed limit and dynamic global optimization; in abnormal situations such as sensor failure and communication interruption, a single decision source may cause speed control failure.

[0006] (3) Simple speed control strategy, difficult to adapt to complex underground working conditions: using preset speed curve or simple reactive obstacle avoidance strategy, unable to dynamically adapt to changes in roadway environment; in complex scenes such as turning, intersection and multi-vehicle intersection, existing speed control methods are prone to sudden acceleration and deceleration; lack of predictive speed planning based on global information, resulting in low vehicle traffic efficiency.

[0007] (4) Insufficient information fusion capability, affecting the quality of speed decision: there is a lack of effective information sharing mechanism between vehicles, roadside facilities and cloud systems; unable to fully utilize the supplementary information of roadside perception devices to enhance the accuracy of speed decision; the complex underground communication environment affects the reliable transmission of speed control instructions due to the instability of single communication method.

[0008] These problems seriously restrict the safety and operation efficiency of unmanned vehicles in underground mines.

[0009] For example, the related technical document CN119723925A discloses a vehicle-road cooperative perception and parking method for a mine shaft environment, which includes: S1, collecting road-side laser radar data, road-side camera data, road-side laser range finder data, vehicle-mounted laser radar data, and vehicle-mounted camera data; S2, obtaining road-side obstacle information according to the collected road-side laser radar data, road-side camera data, and road-side laser range finder data through time synchronization, space synchronization, and calculation of matching factors; S3, obtaining vehicle-mounted obstacle information according to the collected vehicle-mounted laser radar data and vehicle-mounted camera data through time synchronization, space synchronization, and calculation of matching factors; S4, matching vehicle-road obstacles through the Hungarian algorithm according to the road-side obstacle information and the vehicle-mounted obstacle information, and generating a vehicle-road cooperative perception result; S5, performing global path planning by calculating an optimal driving path from a starting point to an unloading point according to the vehicle-road cooperative perception result and using a global map, and generating a global path; S6, performing local path planning through a path obstacle avoidance algorithm according to the vehicle-road cooperative perception result, and generating a local path; and S7, controlling the driving path, speed, and parking of the unmanned vehicle according to the global path and the local path. However, this scheme only relies on vehicle-road two-level cooperation and path planning to control the speed, and cannot foresee and coordinate potential conflicts of multiple vehicles at mine road intersections and turning areas, which is easy to cause vehicle congestion or collision. SUMMARY

[0010] In view of the fact that the speed control of the unmanned vehicle in the mine depends on a single decision source in a narrow tunnel environment, which leads to low traffic efficiency in the intersection and turning blind area of multiple vehicles, the present application provides a mine unmanned vehicle speed control method based on vehicle-tunnel-cloud cooperation, which performs multi-vehicle cooperative control through a three-layer speed control architecture of cloud global expected speed-vehicle end expected speed-final control speed, thereby improving the traffic efficiency.

[0011] One aspect of the present application provides a mine unmanned vehicle speed control method based on vehicle lane cloud cooperation, comprising: collecting vehicle end sensor data, the vehicle end sensor data including laser SLAM data, visual SLAM data, IMU data and wheel speed sensor data; generating vehicle state data including vehicle positioning data and attitude data according to the vehicle end sensor data; collecting vehicle end environment perception data, and generating vehicle end obstacle data set through fusion algorithm according to the vehicle end environment perception data; collecting roadside environment perception data through roadside device, and generating roadside obstacle data set through perception fusion algorithm according to the roadside environment perception data; constructing a global obstacle map according to the vehicle end obstacle data set, the roadside obstacle data set and the vehicle state data, and extracting obstacle data within a preset range with each vehicle position as the center to generate a vehicle obstacle map of each vehicle; calculating the global expected vehicle speed of each vehicle based on the global obstacle map and all vehicle state data to perform multi-vehicle cooperative dynamic control; calculating the vehicle end expected vehicle speed according to the vehicle end obstacle data set, the roadside obstacle data set and the vehicle obstacle map issued by the cloud end; and generating the final control vehicle speed according to the vehicle end expected vehicle speed, the map speed limit and the global expected vehicle speed.

[0012] Further, the vehicle state data including vehicle positioning data and attitude data generated according to the vehicle end sensor data comprises: performing point cloud registration processing on the laser SLAM data to generate vehicle position and attitude estimation based on laser point cloud; performing feature extraction and matching on the visual SLAM data to generate vehicle position and attitude estimation based on visual features; obtaining three-axis acceleration and three-axis angular velocity of the vehicle according to the IMU data; obtaining linear velocity of the vehicle according to the wheel speed sensor data; and performing data fusion by using extended Kalman filtering algorithm to obtain vehicle positioning data and attitude data, wherein the positioning data includes three-dimensional position coordinates of the vehicle in the lane coordinate system, and the attitude data includes yaw angle, pitch angle and roll angle.

[0013] Further, the data fusion by using extended Kalman filtering algorithm to obtain vehicle positioning data and attitude data comprises: in the prediction step, based on the vehicle state at the last moment, the position prediction value at the current moment is calculated by using three-axis acceleration through kinematics equation, and the attitude prediction value at the current moment is calculated by using three-axis angular velocity through attitude update equation; in the update step, the vehicle position and attitude estimation based on laser point cloud, and the vehicle position and attitude estimation based on visual features are taken as observation, and the position prediction value and the attitude prediction value are corrected through Kalman gain; the linear velocity obtained by the wheel speed sensor is compared with the velocity obtained by integrating acceleration, when the difference between the two exceeds a preset threshold, the velocity obtained by integrating acceleration is corrected by using the linear velocity to prevent IMU integral drift; and the corrected three-dimensional position coordinates and attitude angle of the vehicle are output as vehicle positioning data and attitude data.

[0014] In particular, the GPS signal is completely missing in the mine tunnel environment, and the traditional positioning method is invalid. Under the special environment of the mine, single sensors have inherent defects: laser SLAM performance declines in a dusty environment, visual SLAM is limited by the lighting conditions in the mine, IMU has an integral drift problem, and the wheel speed meter is inaccurate when it slips.

[0015] Therefore, the application fuses the data of four sensors of laser SLAM, visual SLAM, IMU, and wheel speed meter, uses an extended Kalman filter algorithm, and improves the positioning accuracy: laser SLAM uses the structured features of the tunnel wall to provide centimeter-level position anchoring, visual SLAM enhances the positioning robustness in areas with rich textures, IMU ensures the continuity of state updates at a high frequency of 200Hz, and the wheel speed meter provides a reliable speed reference to suppress IMU drift.

[0016] In particular, in the typical scenario of a tunnel turn, the traditional scheme often cannot accurately determine the vehicle position until it approaches the bend due to large positioning errors, resulting in frequent emergency braking. The application improves the positioning accuracy to lay the foundation for subsequent vehicle speed control.

[0017] Further, the vehicle-end environmental perception data is collected, and the vehicle-end obstacle data set is generated according to the vehicle-end environmental perception data through a fusion algorithm, including: collecting vehicle-end environmental perception data, the vehicle-end environmental perception data including laser radar point cloud data, millimeter wave radar data, and camera image data; according to the laser radar point cloud data, by filtering out the static point cloud of the mine tunnel roof and side wall, the dynamic point cloud is obtained, the dynamic point cloud is clustered, and the three-dimensional contour data of the obstacle is obtained; according to the millimeter wave radar data, the radial velocity and distance of the obstacle detected by the millimeter wave radar are extracted, and the speed of the obstacle relative to the vehicle is calculated according to the Doppler effect; according to the camera image data, the obstacle category is identified through a target detection algorithm; the three-dimensional contour data, the relative speed of the obstacle, and the obstacle category are time-aligned and space-registered through a space-time synchronization algorithm, and a multi-sensor fusion algorithm is used to generate the vehicle-end obstacle data set; the vehicle-end obstacle data set is associated with the vehicle state data, transmitted to surrounding vehicles and roadside devices through V2X communication, and transmitted to a cloud server through 5G communication;

[0018] Further, a multi-sensor fusion algorithm is used to generate the vehicle-end obstacle dataset, including: taking three-dimensional contour data, obstacle relative speed and obstacle category as input, and estimating the obstacle state through an extended Kalman filter algorithm; according to a preset obstacle category library, the identified obstacle is matched to one of the following categories: mine car, roadway worker, fixed equipment, temporary obstacle or other moving object; the obstacle distance and azimuth angle measured by the laser radar are converted into three-dimensional coordinates through coordinate conversion, and the distance and azimuth angle measured by the millimeter wave radar are converted into three-dimensional coordinates, and the obstacle center point position is calculated through a weighted average algorithm, wherein the weight coefficient is adjusted according to the laser radar point cloud density; the radial velocity measured by the millimeter wave radar is decomposed into three-dimensional velocity components; according to the position difference and time interval of the same obstacle in two adjacent laser radar data, the speed estimation value of the obstacle is calculated; the three-dimensional velocity components and the speed estimation value are fused through Kalman filtering to obtain the three-dimensional velocity vector of the obstacle; principal component analysis is performed on the obstacle point cloud obtained by the laser radar, and the principal axis direction is extracted as the obstacle orientation, and the length, width and height of the obstacle are calculated through the minimum circumscribed rectangle algorithm; the distance deviation between the laser radar detection position, the millimeter wave radar detection position and the camera detection bounding box center position is calculated, and the confidence of the corresponding obstacle is calculated according to the distance deviation; the category identification, three-dimensional position coordinates, three-dimensional velocity vector, length, width and height data, orientation angle and confidence of each obstacle are combined to obtain the vehicle-end obstacle dataset.

[0019] In particular, the narrow roadway space makes the obstacle close to the vehicle and the reaction time short, dust and water mist seriously affect the sensor performance, and the dim lighting conditions limit the visual perception ability, and the strong reflection of the roadway wall surface easily causes radar false detection.

[0020] Therefore, in this application, the laser radar focuses on dynamic obstacle detection by filtering out the static point cloud of the roadway roof and side wall, and its three-dimensional point cloud clustering algorithm can accurately obtain the three-dimensional contour of the obstacle, even irregularly shaped temporary obstacles can be accurately identified. The millimeter wave radar directly measures the radial velocity of the obstacle using the Doppler effect, which is not affected by light and dust, and provides a reliable reference for the speed estimation of moving obstacles. The camera identifies the obstacle category through the deep learning target detection algorithm, and accurately classifies it into mine cars, workers, fixed equipment, temporary obstacles and the like, providing semantic information for subsequent speed decision-making.

[0021] In addition, the scheme realizes the deep fusion of the data of the three sensors through extended Kalman filtering. In the spatial dimension, the position of the obstacle is accurately calculated through coordinate conversion and weighted average algorithm, and the weight coefficient is dynamically adjusted according to the point cloud density of the laser radar, so that the optimal position estimation can be obtained under different distances and shielding conditions. In the speed dimension, the radial speed measurement of the millimeter wave radar and the speed estimation calculated by the inter-frame displacement of the laser radar are Kalman filtered and fused to obtain an accurate three-dimensional speed vector. This fusion is not a simple data superposition, but makes full use of the advantages of each sensor: when the dust affects the laser radar, the speed measurement of the millimeter wave radar ensures the reliable detection of the dynamic obstacle; when the millimeter wave radar produces false alarm due to multipath effect, the accurate profile of the laser radar can be effectively eliminated; when both kinds of radars are difficult to judge the type of the obstacle, the visual recognition provides key semantic information.

[0022] Further, the roadside environment perception data is collected by the roadside device, and the roadside obstacle data set is generated by the perception fusion algorithm according to the roadside environment perception data, including: arranging the roadside device at the turning place, intersection place and straight roadway of the mine roadway according to the preset interval, the roadside device is configured with a laser radar, a millimeter wave radar and a camera; collecting roadside environment perception data through the roadside device, the roadside environment perception data includes roadside laser radar point cloud data, roadside millimeter wave radar data and roadside camera image data; according to the roadside laser radar point cloud data, the static point cloud of the roadway roof, sidewall and ground is filtered to obtain dynamic point cloud, the dynamic point cloud is clustered to obtain three-dimensional profile data of the obstacle relative to the roadside device; according to the roadside millimeter wave radar data, the radial speed and distance of the obstacle detected by the roadside millimeter wave radar are extracted, and the speed of the obstacle relative to the roadside device is calculated; according to the roadside camera image data, the obstacle category is identified through a target detection algorithm, and the obstacle is matched to one of the mine car, the roadway operation personnel, the fixed equipment, the temporary obstacle or other moving object category; the three-dimensional profile data, the relative speed of the obstacle and the obstacle category obtained by the roadside device are time-aligned and space-registered through a space-time synchronization algorithm; the perception data of the roadside device is processed by using the same multi-sensor fusion algorithm as the vehicle-end obstacle data set to generate the roadside obstacle data set containing the obstacle category identification, the three-dimensional position coordinates relative to the roadside device, the three-dimensional speed vector, the length-width-height data, the orientation angle and the detection confidence; the roadside obstacle data set is associated with the position information of the roadside device, and is sent to the unmanned vehicle within the detection range through V2X communication, and is sent to the cloud server through 5G communication or wired network.

[0023] Further, the vehicle obstacle map of each vehicle is generated, including: the cloud receiving the vehicle end obstacle data set and the vehicle state data sent by the plurality of vehicles through 5G communication, and the roadside obstacle data set sent by the plurality of roadside devices through 5G communication or wired network; converting the three-dimensional position coordinates in the vehicle end obstacle data set relative to the vehicle into global coordinates under the lane coordinate system according to the positioning data of the corresponding vehicle; converting the three-dimensional position coordinates in the roadside obstacle data set relative to the roadside device into global coordinates under the lane coordinate system according to the preset position of the roadside device; matching the same obstacle from different data sources: calculating the Euclidean distance of the global coordinates of the obstacles in different data sources, and determining that they are the same obstacle when the Euclidean distance is less than a preset threshold; fusing the same obstacle data matched: obtaining the obstacle category, length, width, height and orientation angle corresponding to the data source with the highest confidence, and performing weighted average on the position coordinates and velocity vectors of multiple data sources; constructing a global obstacle map according to the fused obstacle data; each obstacle in the global obstacle map includes: obstacle category identifier, global three-dimensional position coordinates, global three-dimensional velocity vector, length, width, height, orientation angle and confidence; according to the vehicle positioning data of each vehicle, the obstacle data within a preset range is extracted with the vehicle position as the center, and the vehicle obstacle map of the corresponding vehicle is generated; and sending the vehicle obstacle map of each vehicle to the corresponding vehicle through 5G communication;

[0024] In particular, one of the biggest challenges in the mine tunnel environment is the severe limitation of the single vehicle's perception field of view. The tunnel is narrow and winding, and there are a large number of perception blind spots at turning positions, T-shaped intersections and other positions of the vehicle-mounted sensor, which cannot grasp the global traffic situation of the tunnel. This limitation directly affects the optimization space of speed control - the vehicle can only passively adjust the speed based on local information, and it is difficult to achieve predictive speed planning.

[0025] Therefore, the cloud end of the present application collects all vehicle end obstacle data sets and all roadside obstacle data sets of roadside devices, and realizes all-around perception coverage of the tunnel environment. Each vehicle and each roadside device becomes a perception node of the entire system, and their respective local perception information is integrated into a complete global view in the cloud. First, all obstacle data is converted to a unified lane coordinate system, then the same obstacle from different data sources is identified through Euclidean distance matching, and finally the most accurate obstacle information is obtained based on confidence weighted fusion. This fusion is not a simple data splicing, but makes full use of the complementarity of multi-view observation - the vehicle provides dynamic updates from a moving perspective, and the roadside device provides stable observations from a fixed perspective, which greatly improves the integrity and accuracy of obstacle detection.

[0026] Further, the cloud calculates the global expected speed of each vehicle based on the global obstacle map and all vehicle state data to perform multi-vehicle cooperative dynamic control, including: obtaining obstacle data within a predetermined distance of the driving direction of each vehicle from the global obstacle map, calculating the safe speed required to avoid collision according to the relative position and relative speed of the obstacle and the vehicle; identifying intersections and turning positions in the tunnel and detecting whether there is a path conflict situation in which multiple vehicles approach the same intersection or curve; when detecting that multiple vehicles have path conflicts, determining the vehicle passing priority according to the following rules: calculating the expected time of each vehicle to reach the conflict point; determining the priority weight according to the vehicle type, load state and driving direction; generating a vehicle passing sequence according to the priority weight; for vehicles with low priority, calculating the corresponding deceleration according to the distance of the corresponding vehicle from the conflict point and the current vehicle speed to generate the global expected speed after deceleration; for vehicles with high priority, the safe speed calculated based on obstacle avoidance is used as the global expected speed; for vehicles without path conflicts, the safe speed calculated based on obstacle avoidance is used as the global expected speed; and sending the global expected speed of each vehicle to the corresponding vehicle through 5G communication.

[0027] In particular, the width of the mine tunnel is usually only 3-4 meters, allowing only single vehicle passage, and multiple vehicle conflicts are easily formed at key nodes such as intersections and turns. In the traditional single vehicle independent decision-making mode in this environment: each vehicle only controls the speed based on its own perception, lacks global coordination, and when multiple vehicles meet in narrow sections, they can only avoid collision through emergency braking and inefficient waiting, seriously affecting transportation efficiency and safety. The present scheme calculates the differentiated global expected speed of each vehicle based on global information by the cloud, fundamentally solving the speed optimization problem of multi-vehicle cooperative operation in mines.

[0028] In addition, the present scheme establishes an intelligent priority allocation mechanism to achieve the optimal passing strategy in conflict situations. Priority is not simply first-come, first-served, but takes into account multiple factors such as vehicle type (e.g., heavy-load transport vehicles have priority over empty vehicles), load state (fully loaded vehicles have a longer braking distance and should have higher priority), and driving direction (vehicles on an uphill have difficulty starting and should have priority). Based on this multi-dimensional priority evaluation, the system maintains normal speed for vehicles with high priority and calculates a reasonable deceleration curve for vehicles with low priority. This differentiated speed control ensures overall passing efficiency while avoiding frequent start-stop operations.

[0029] Finally, the cloud performs optimal speed planning based on complete global information, solving the problem of limited single-vehicle vision that cannot perform global optimization; at the same time, the global expected speed is sent to the vehicle end as a guide rather than a mandatory instruction, preserving the ability of the vehicle end to make safety adjustments based on real-time perception. This hierarchical cooperative mechanism achieves global optimization while ensuring local safety.

[0030] Further, the vehicle end desired vehicle speed is calculated, including: obtaining the vehicle end obstacle data set generated by the vehicle end self perception, the obstacle data set sent by other vehicles received through V2X communication, the roadside obstacle data set sent by the roadside device received through V2X communication, and the vehicle obstacle map issued by the cloud received through 5G communication; converting all obstacle information to the vehicle coordinate system, correlating the data according to the position of the obstacle, and judging whether the obstacles in different data sources are the same target; fusing the obstacle data correlated to the same target: when multiple data sources detect the same obstacle, selecting the obstacle category of the data source with the highest confidence; performing weighted average on the position and speed of the obstacle; retaining the independent obstacles detected in all data sources; selecting the effective obstacle closest to the driving path, calculating the relative distance and relative speed of the effective obstacle, and determining the corresponding vehicle end desired vehicle speed through a preset query table, which presets the safe vehicle speed under different distance and speed combinations in the mine environment; for the distance and speed combinations not covered in the query table, a linear interpolation method is used to calculate the vehicle end desired vehicle speed.

[0031] Further, the final control vehicle speed is generated, including: obtaining the vehicle end desired vehicle speed, the map speed limit of the lane road section where the current vehicle is located, and the global desired vehicle speed sent by the cloud, selecting the minimum value among the three as the preliminary control vehicle speed; obtaining the final control vehicle speed of the last control period; calculating the difference between the preliminary control vehicle speed and the final control vehicle speed of the last control period; when the difference is greater than the product of the preset acceleration threshold and the control period length, the current control vehicle speed , wherein, is the current control vehicle speed; is the final control vehicle speed of the last control period; is the preliminary control vehicle speed; sign() is a sign function, returning +1 when the value in the parentheses is positive, and -1 when it is negative; is the maximum allowed acceleration; T is the control period length.

[0032] When the difference is less than or equal to the product of the preset acceleration threshold and the control period length, the current control vehicle speed , wherein, a is a smoothing coefficient, with a value range of 0 to 1; the output current control vehicle speed is the final control vehicle speed of the current control period.

[0033] In particular, the biggest challenge for unmanned mining vehicles is how to ensure safe operation under various abnormal conditions such as communication interruption, sensor failure, and sudden obstacles. If completely dependent on cloud instructions, the vehicle will lose control once 5G communication is interrupted; if only relying on on-board sensing, it cannot obtain globally optimized speed guidance.

[0034] Therefore, in this application, the vehicle not only relies on its own sensors for perception, but also receives perception information from surrounding vehicles and roadside devices through V2X, and obtains a global obstacle map from the cloud through 5G, forming a triple perception system of autonomous perception, cooperative perception, and global perception. When the vehicle's own laser radar is blocked by dust, it can rely on other vehicle perception data obtained through V2X; when entering a turning blind area, the perception information of roadside devices provides a key supplement; even if V2X communication is disturbed, the obstacle map issued by the cloud can still provide basic protection. Through confidence weighting and data association, the system can intelligently extract the most reliable obstacle information from multiple data sources, significantly improving the integrity and accuracy of perception.

[0035] In addition, traditional real-time calculation methods are prone to calculation delay or abnormality in complex environments, while the present scheme pre-establishes a safe speed lookup table under different distance and speed combinations through a large amount of actual operation data of mine vehicles, avoiding complex real-time calculation and greatly improving response speed and reliability.

[0036] Finally, mine vehicles usually have large load capacity, and sudden acceleration and deceleration not only affect the safety of goods, but also cause huge energy consumption and equipment wear and tear. This scheme ensures the smoothness of speed change through acceleration limitation and exponential smoothing algorithm.

[0037] Another aspect of the present application also provides a mine unmanned vehicle speed control system based on vehicle-lane cloud cooperation, comprising: a vehicle-mounted intelligent device, which collects vehicle end sensor data and generates vehicle state data containing vehicle positioning data and attitude data; collects vehicle end environment perception data and generates vehicle end obstacle data set through fusion algorithm; receives obstacle data sent by other vehicles, roadside devices and cloud, and calculates vehicle end expected speed after data fusion; generates the final control vehicle speed according to the vehicle end expected speed, map speed limit and global expected speed; a roadside intelligent device, which is arranged at the turning place, intersection place and preset position of straight lane of the mine roadway, and is configured with laser radar, millimeter wave radar and camera; is used for collecting roadside environment data, generating roadside obstacle data set through perception fusion algorithm, sending the roadside obstacle data set to the vehicle through V2X communication, and sending the roadside obstacle data set to the cloud processing device through 5G communication or wired network; a mine communication base station, which provides 5G communication link between the vehicle-mounted intelligent device and the cloud processing device, and provides V2X communication link between the vehicle-mounted intelligent devices and between the vehicle-mounted intelligent device and the roadside intelligent device; a cloud processing device, comprising a data fusion server, a map construction server and a cooperative decision server; is used for receiving vehicle end obstacle data set of multiple vehicles, vehicle state data and roadside obstacle data set of multiple roadside devices; constructing a global obstacle map, and generating a vehicle obstacle map for each vehicle; based on the global obstacle map and all vehicle state data, calculating the global expected speed of each vehicle for multi-vehicle cooperative control.

[0038] Compared with the prior art, the present application has the following advantages:

[0039] In the present application, the global expected speed of each vehicle is calculated based on the global obstacle map in the cloud, which can identify and coordinate potential conflicts of multiple vehicles at key positions such as lane intersections and turning in advance. For example, when two vehicles approach a T-shaped intersection at the same time, the cloud can issue a speed reduction instruction to one of the vehicles according to the vehicle arrival time and priority, avoiding dangerous situations of emergency braking at the last moment relying only on vehicle-mounted perception.

[0040] On the other hand, the three-layer speed control architecture realizes the organic combination of global optimization and local control. The cloud can make predictive speed planning based on global traffic situation, and the vehicle end can make dynamic adjustment according to real-time perception. In the actual application of the main transportation lane of the mine, through the advance scheduling of the cloud, the invalid waiting of multiple vehicles in the narrow section is avoided, and the vehicle passing efficiency is improved.

[0041] In addition, through the multiple constraint mechanism of the vehicle end expected vehicle speed, the map speed limit, and the global expected vehicle speed, the system can still operate safely even in the case of partial sensor failure or communication interruption. For example, when the 5G communication is temporarily interrupted and the cloud global expected vehicle speed cannot be obtained, the vehicle can still be controlled safely based on the vehicle end perception and the map speed limit; when the vehicle-mounted sensor fails, the global expected vehicle speed and the map speed limit issued by the cloud can be relied on to ensure basic safety. The parking rate caused by system failure is reduced, and the traffic efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0042] The present application will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same numbers represent the same structures, wherein:

[0043] Figure 1 is an exemplary block diagram of a mine unmanned vehicle speed control system based on vehicle-lane-cloud collaboration according to some embodiments of the present application;

[0044] Figure 2 is an exemplary flowchart of a mine unmanned vehicle speed control method based on vehicle-lane-cloud collaboration according to some embodiments of the present application;

[0045] Figure 3 is a schematic diagram of a fusion positioning algorithm according to some embodiments of the present application;

[0046] Figure 4 is a schematic diagram of a fusion perception algorithm according to some embodiments of the present application;

[0047] Figure 5 is a schematic diagram of an obstacle map construction according to some embodiments of the present application;

[0048] Figure 6 is an exemplary flowchart of vehicle end data fusion according to some embodiments of the present application. DETAILED DESCRIPTION

[0049] The method and system provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0050] Embodiment 1

[0051] As shown in Figure 1 , a mine unmanned vehicle speed control system based on vehicle-lane-cloud collaboration includes:

[0052] The vehicle-mounted intelligent device includes a vehicle-mounted domain controller, a camera, a laser radar, a V2X communication module, and a millimeter wave radar. Vehicle sensor data is collected and vehicle state data including vehicle positioning data and attitude data is generated. Vehicle environment perception data is collected and a vehicle obstacle data set is generated through a fusion algorithm. Obstacle data sent by other vehicles, roadside devices, and the cloud is received, and after data fusion, the expected vehicle speed at the vehicle end is calculated. The final control vehicle speed is generated according to the expected vehicle speed at the vehicle end, the map speed limit, and the global expected vehicle speed.

[0053] The roadside intelligent device is arranged at a turning, an intersection, and a preset position of a straight tunnel of a mine, and is configured with a roadside computing unit, a camera, a laser radar, a V2X communication module, and a millimeter wave radar. The roadside intelligent device is used for collecting roadside environment data, generating a roadside obstacle data set through a perception fusion algorithm, sending the roadside obstacle data set to vehicles through V2X communication, and sending the roadside obstacle data set to a cloud processing device through 5G communication or a wired network.

[0054] The underground communication base station provides a 5G communication link between the vehicle-mounted intelligent device and the cloud processing device, and provides a V2X communication link between the vehicle-mounted intelligent devices and between the vehicle-mounted intelligent device and the roadside intelligent device.

[0055] The cloud processing device includes a data fusion server, a map construction server, a collaborative decision-making server, and a communication gateway. The cloud processing device is used for receiving vehicle obstacle data sets of multiple vehicles, vehicle state data, and roadside obstacle data sets of multiple roadside devices, constructing a global obstacle map, generating vehicle obstacle maps of the vehicles, calculating global expected vehicle speeds of the vehicles based on the global obstacle map and all vehicle state data, and performing multi-vehicle collaborative control.

[0056] Embodiment 2

[0057] As Figure 2As shown, the vehicle end sensor data is collected, including laser SLAM data, visual SLAM data, IMU data and wheel speed data; vehicle state data including vehicle positioning data and attitude data is generated according to the vehicle end sensor data; vehicle end environment perception data is collected, and a vehicle end obstacle data set is generated through a fusion algorithm according to the vehicle end environment perception data; roadside environment perception data is collected through a roadside device, and a roadside obstacle data set is generated through a perception fusion algorithm according to the roadside environment perception data; the cloud end constructs a global obstacle map according to the vehicle end obstacle data set, the roadside obstacle data set and the vehicle state data, and extracts obstacle data within a preset range centered on the position of each vehicle to generate a vehicle obstacle map for each vehicle; the cloud end calculates the global expected vehicle speed of each vehicle based on the global obstacle map and all vehicle state data for multi-vehicle cooperative dynamic control; the vehicle end calculates the vehicle end expected vehicle speed according to the vehicle end obstacle data set, the roadside obstacle data set and the vehicle obstacle map issued by the cloud end; and the vehicle end generates the final control vehicle speed according to the vehicle end expected vehicle speed, the map speed limit and the global expected vehicle speed.

[0058] Specifically, as shown in the figure, Figure 3 The system receives the original point cloud data obtained by laser radar scanning, performs point cloud registration through the iterative closest point (ICP) algorithm, and generates the three-dimensional position and attitude angle of the vehicle relative to the map coordinate system.

[0059] The image data collected by the camera first passes through the ORB feature extraction algorithm to detect the key points in the image and calculate the feature descriptors. The current frame is matched with the key frame through the feature matching algorithm, and the RANSAC algorithm is used to remove the false matching points. Based on the matched feature point pairs, the camera pose is solved through the PnP algorithm, and then the position and attitude of the vehicle are converted according to the calibration relationship between the camera and the vehicle body.

[0060] The IMU sensor outputs original measurement data at a frequency of 200Hz, including three-axis acceleration and three-axis angular velocity . The data is filtered by a low-pass filter to remove high-frequency noise, and is corrected according to the pre-calibrated bias and scale factor.

[0061] The wheel speed sensor outputs the wheel speed signal, and the longitudinal linear velocity of the vehicle is calculated according to the tire radius and the transmission ratio. Considering the possible differences between the left and right wheels, the average of the two wheel speeds is taken as the vehicle speed estimate.

[0062] The system state vector is defined as , including 9 state quantities of position, velocity and attitude.

[0063] Prediction step:

[0064] Position prediction: according to the state at the last time step and IMU acceleration measurements, the position prediction at the current time step is calculated by kinematic equations:

[0065] ;

[0066] ;

[0067] ;

[0068] Velocity prediction: , Similarly.

[0069] Attitude prediction: using angular velocity to calculate by quaternion or Euler angle update equation:

[0070] ; ; ;

[0071] Covariance prediction: where F is the state transition matrix, Q is the process noise covariance matrix, is the covariance matrix at the last time step.

[0072] Update step:

[0073] Observation vector construction: combine the position and attitude estimates of laser SLAM and visual SLAM into an observation vector Kalman gain calculation: where H is the observation matrix, R is the observation noise covariance matrix

[0074] State update: ; Covariance update: .

[0075] Velocity correction mechanism:

[0076] Calculate the IMU integrated velocity: , denotes the acceleration provided by the IMU; denotes the speed of the vehicle at the last time step; denotes the integrated velocity calculated by the IMU.

[0077] Velocity difference calculation: where, denotes the linear velocity obtained by the wheel speed sensor.

[0078] Condition: if Δv > threshold (e.g. 0.5 m / s)

[0079] Correction execution: where, is the correction coefficient (usually 0.7-0.9), represents the corrected speed.

[0080] The final output includes: vehicle three-dimensional position coordinates: (x, y, z), representing the position of the vehicle in the tunnel coordinate system; vehicle attitude angle: (roll, pitch, yaw), representing the roll angle, pitch angle, and heading angle of the vehicle; output frequency: 100 Hz, fixed frequency output is achieved through timestamp interpolation; data format: packaged according to standard message format, containing timestamp, position, attitude, and covariance information.

[0081] As shown in Figure 4 , collect multi-modal sensor data and preprocess, laser radar outputs raw point cloud data at a frequency of 10 Hz, each frame contains about 100,000 points, each point contains (x, y, z, intensity) information. Based on the RANSAC algorithm, fit the tunnel roof and side wall planes, mark and remove points with a distance less than a threshold (usually 0.1 m) from the fitted plane as static points; keep the remaining point cloud as potential dynamic targets, downsample the point cloud to a resolution of 0.1 m through a voxel grid. Use the Euclidean distance clustering algorithm, set the clustering radius to 0.3 m and the minimum number of points to 10, segment the point cloud into independent obstacle clusters, calculate the convex hull for each cluster to obtain the three-dimensional contour vertex coordinate sequence of the obstacle.

[0082] Radar output includes target ID, radial distance r, azimuth angle θ, elevation angle , radial velocity , reflection intensity, etc. Convert spherical coordinates to Cartesian coordinates: , , . Based on the Doppler effect, radial velocity is only the projection component of the velocity in the radial direction, and the complete velocity needs to be estimated in combination with the target motion model.

[0083] Distortion correction and size normalization (640x480) are performed on the original RGB image, YOLOv5 algorithm is used, and output includes bounding box , class ID, and confidence score; map the detected class ID to the predefined class: 1 - mine car, 2 - worker, 3 - fixed equipment, 4 - temporary obstacle, 5 - other.

[0084] Record the time stamp of each sensor data based on the system clock, maintain a 200ms cycle buffer for each sensor, and find the nearest millimeter wave and image data within the [t-50ms, t+50ms] window for the laser radar trigger time t. Time interpolation is performed on the millimeter wave radar data to ensure alignment with the laser radar time.

[0085] Convert all sensor data to the vehicle body coordinate system with the origin at the center of the rear axle, convert according to the installation parameter matrix T_laser_to_vehicle, and apply the calibration matrix T_radar_to_vehicle; project the image detection frame to the three-dimensional space through the intrinsic matrix K and the extrinsic matrix [R|t].

[0086] Calculate the Euclidean distance of the target detected by different sensors. When the distance is less than 1.5m and the time difference is less than 50ms, it is determined as the same target; construct an MxN association matrix, where M is the number of laser targets and N is the number of radar targets.

[0087] Use extended Kalman filter for state estimation:

[0088] State vector definition: the state vector of each obstacle contains 6 components of position and velocity.

[0089] Prediction step:

[0090] State prediction: where F is the state transition matrix of the constant speed motion model;

[0091] Covariance prediction: where Q is the process noise;

[0092] Update step:

[0093] Observation construction: ;

[0094] Kalman gain: ;

[0095] State update: ;

[0096] Multi-sensor fusion calculation

[0097] Position fusion: where ρ is the point cloud density, is the reference density (100 points / m³); Uncertainty estimation: calculate the covariance of the fused position according to the sensor measurement variance.

[0098] Velocity fusion: millimeter wave velocity decomposition: decompose the radial velocity ​Decomposed into three-dimensional components; laser inter-frame velocity: , represents the position of the current frame point cloud, represents the position of the previous frame point cloud, Kalman filter: optimally fuse the two kinds of velocity estimates as observations.

[0099] Geometric feature extraction: principal component analysis (PCA): calculate the covariance matrix of the obstacle point cloud, and the eigenvector corresponds to the principal axis direction; orientation angle calculation: yaw=atan2 (principal axis y component, principal axis x component); size calculation: project the point cloud along the principal axis direction, and calculate the minimum circumscribed rectangle to obtain length, width and height; confidence calculation: spatial consistency: , represents the position of the laser radar, represents the position of the millimeter wave radar, represents the position of the camera; confidence score: , σ is the standard deviation parameter.

[0100] Combine the category identification, three-dimensional position coordinates, three-dimensional velocity vector, length, width, height data, orientation angle and confidence of each obstacle to obtain the vehicle-end obstacle data set. Associate the vehicle-end obstacle data set with the vehicle state data, send it to the surrounding vehicles and roadside devices through V2X communication, and send it to the cloud server through 5G communication;

[0101] The roadside environment perception data is collected by a roadside device, and a roadside obstacle data set is generated by a perception fusion algorithm according to the roadside environment perception data, comprising: arranging roadside devices at turning places, intersection places and straight roadways of the mine roadway according to a preset interval, the roadside device being provided with a laser radar, a millimeter wave radar and a camera; collecting roadside environment perception data by the roadside device, the roadside environment perception data comprising roadside laser radar point cloud data, roadside millimeter wave radar data and roadside camera image data; obtaining dynamic point cloud by filtering out static point cloud of the roadway roof, sidewall and ground according to the roadside laser radar point cloud data, clustering the dynamic point cloud to obtain three-dimensional contour data of the obstacle relative to the roadside device; extracting radial velocity and distance of the obstacle detected by the roadside millimeter wave radar according to the roadside millimeter wave radar data, and calculating the speed of the obstacle relative to the roadside device; identifying the obstacle category by a target detection algorithm according to the roadside camera image data, and matching the obstacle to one of a mine car, a roadway operator, a fixed device, a temporary obstacle or other moving object category; performing time alignment and space registration on the three-dimensional contour data, the relative speed of the obstacle and the obstacle category obtained by the roadside device by a time-space synchronization algorithm; processing the perception data of the roadside device by using the same multi-sensor fusion algorithm as the vehicle-end obstacle data set to generate a roadside obstacle data set containing obstacle category identification, three-dimensional position coordinates relative to the roadside device, three-dimensional velocity vector, length-width-height data, orientation angle and detection confidence; associating the roadside obstacle data set with the position information of the roadside device, and sending it to the unmanned vehicle within the detection range through V2X communication, and sending it to the cloud server through 5G communication or wired network.

[0102] As shown in Figure 5 , the cloud receives data packets uploaded by each vehicle through the 5G network, and the data format contains vehicle ID, timestamp, vehicle-end obstacle data set and vehicle state data (position, speed, heading); the roadside device data is received through 5G or wired network, containing device ID, timestamp and roadside obstacle data set.

[0103] For the obstacle j detected by the vehicle i, the global coordinate calculation formula is: ; wherein: is a rotation matrix based on the vehicle heading angle; is the position vector of the vehicle in the roadway coordinate system.

[0104] The obstacle detected by the roadside device k is converted according to the pre-calibrated device position and orientation: ; wherein: is the fixed transformation parameter of the roadside device.

[0105] For each obstacle, the Euclidean distance of the obstacle to other data source obstacles is calculated, wherein P is the global position of the obstacle, when When the distance is < 2.0 m (a typical value) and the time difference is < 100 ms, it is determined that the same target is preferred. A Hungarian algorithm is used to solve the one-to-many matching conflict.

[0106] For the same obstacle matched, the data sources are arranged in descending order of confidence, and the obstacle class, length, width, height and orientation angle corresponding to the data source with the highest confidence are obtained. The position coordinates and velocity vectors of multiple data sources are weighted and averaged, and the weight is proportional to the confidence: The fused obstacle data is constructed into a global obstacle map according to the global coordinates. Each obstacle in the global obstacle map includes: obstacle class identifier, global three-dimensional position coordinates, global three-dimensional velocity vector, length, width, height, orientation angle and confidence. According to the vehicle positioning data of each vehicle, the obstacle data within a predetermined range (for example, 30-50 meters) centered on the vehicle position is extracted to generate a vehicle obstacle map corresponding to the vehicle. Each vehicle obstacle map is sent to the corresponding vehicle through 5G communication.

[0107] The obstacle data within a predetermined distance in the driving direction of each vehicle is obtained from the global obstacle map. According to the relative position and relative speed of the obstacle and the vehicle, the safe vehicle speed required to avoid collision is calculated. The intersection and turning position in the lane are identified, and whether there is a path conflict situation in which multiple vehicles approach the same intersection or curve at the same time is detected. When multiple vehicles are detected to have a path conflict, the vehicle passing priority is determined according to the following rules: the expected time of each vehicle to reach the conflict point is calculated; the priority weight is determined according to the vehicle type, load state and driving direction; the vehicle passing order sequence is generated according to the priority weight; for vehicles with low priority, the corresponding deceleration is calculated according to the distance of the corresponding vehicle from the conflict point and the current vehicle speed, and the global expected vehicle speed after deceleration is generated; for vehicles with high priority, the safe vehicle speed calculated based on obstacle avoidance is used as the global expected vehicle speed; for vehicles without path conflict, the safe vehicle speed calculated based on obstacle avoidance is used as the global expected vehicle speed; the global expected vehicle speed of each vehicle is sent to the corresponding vehicle through 5G communication.

[0108] As Figure 6As shown, the vehicle's own perception system generates a local obstacle dataset at a frequency of 10 Hz, containing obstacle information fused from lidar, millimeter-wave radar, and camera. Obstacle data broadcast by other vehicles within a range of 300 meters is received through V2X short-range communication (DSRC / C-V2X), with the receiving frequency depending on the number of neighboring vehicles, typically receiving 5-10 data packets per second. Roadside devices broadcast obstacle information within their perception range through V2X at a fixed period (usually 100 ms). The cloud issues vehicle obstacle map updates through the 5G network at a lower frequency (2 Hz), but with a wider coverage, including long-range obstacles that cannot be directly perceived by on-board sensors.

[0109] Different data sources use a unified obstacle description format, including fields such as obstacle ID, category, position, speed, size, orientation, confidence, and timestamp. The system maintains independent receiving buffers for each data source, and handles the time delay differences of different data sources through a timestamp alignment mechanism.

[0110] All obstacle data needs to be converted to the vehicle coordinate system with the center of the rear axle as the origin. For other vehicle obstacle data received through V2X, a transformation matrix is first constructed based on the relative position and orientation of the sending vehicle, and then the obstacle coordinates are converted from the sending vehicle coordinate system to the receiving vehicle coordinate system. The conversion of roadside device data requires the use of the vehicle's own positioning and the fixed position information of the roadside device. The obstacle map issued by the cloud has been preprocessed and directly contains coordinate information relative to the receiving vehicle.

[0111] The data association uses a distance-based clustering method. For all obstacle data within a time window (usually 100 ms), their spatial distances in the vehicle coordinate system are calculated. When the Euclidean distance between two obstacles is less than the association threshold (dynamically adjusted according to positioning accuracy, usually 1.5-2.5 meters), it is determined as different observations of the same target. The association process also considers the motion consistency of obstacles, further verifying the correctness of association by comparing the direction and size of the velocity vector.

[0112] Fusion of obstacle data associated with the same target: when multiple data sources detect the same obstacle, the obstacle category of the data source with the highest confidence is selected; the position and speed of the obstacle are weighted averaged; all independent obstacles detected in all data sources are retained; obstacles not associated with are retained as independent targets. The system assigns a temporary ID to each independent obstacle and continues to track it in subsequent frames, avoiding target loss due to temporary detection failure.

[0113] According to the current driving direction of the vehicle, a sector region of interest (ROI) is defined. The parameters of the region include: forward distance (usually 100-150 meters), lateral width (lane width plus safety margin), and vertical height (considering the height limit of the lane). Only the obstacles located in the ROI are considered as valid obstacles.

[0114] The closest valid obstacle on the driving path is selected, the relative distance and relative speed of the valid obstacle are calculated, and the corresponding vehicle-end expected vehicle speed is determined through a preset query table, such as Table 1, which presets the safe vehicle speed under different distance and speed combinations in the underground mine environment; for distance and speed combinations not covered in the query table, a linear interpolation method is used to calculate the vehicle-end expected vehicle speed;

[0115] Table 1: Expected vehicle speed query table (km / h)

[0116]

[0117] The vehicle-end generates the final control vehicle speed according to the vehicle-end expected vehicle speed, the map speed limit, and the global expected vehicle speed, including: obtaining the vehicle-end expected vehicle speed, the map speed limit of the lane section where the current vehicle is located, and the global expected vehicle speed sent by the cloud, selecting the minimum value among the three as the preliminary control vehicle speed; obtaining the final control vehicle speed of the last control period; calculating the difference between the preliminary control vehicle speed and the final control vehicle speed of the last control period; when the difference is greater than the product of the preset acceleration threshold and the control period length, the current control vehicle speed , , where is the current control vehicle speed; is the final control vehicle speed of the last control period; is the preliminary control vehicle speed; sign() is the sign function, returning +1 when the value inside the parentheses is positive and -1 when it is negative; is the maximum allowed acceleration; T is the control period length. When the difference is less than or equal to the product of the preset acceleration threshold and the control period length, the current control vehicle speed , , where α is the smoothing coefficient, taking a value in the range of 0 to 1; the output current control vehicle speed is the final control vehicle speed of the current control period. Through the closed-loop control algorithm, the throttle and brake of the vehicle are controlled to make the vehicle follow the expected vehicle speed.

[0118] The above description of the application and its embodiments is illustrative and not restrictive. Many variations of the application incorporating modifications without departing from the spirit or essential characteristic of the application can now be made by those skilled in the art. For example, the above-described embodiments of the application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Thus, the presently disclosed application is not intended to be limited to the embodiments described herein but is to be accorded the widest scope of the appended claims, the full scope of which is set forth in the appended claims. Furthermore, the terms "comprises", "comprising", "includes", "including", "contains", "containing" or variations thereof do not have a limiting meaning, and do not exclude other elements or steps. The terms "one", "a" or "an" do not exclude a plurality. The terms "first", "second" or the like do not imply any particular order, but are used for naming purposes only.

Claims

1. A well mine unmanned vehicle speed control method based on vehicle lane cloud cooperation, characterized in that, The method comprises the following steps: Collecting vehicle-end sensor data, including laser SLAM data, visual SLAM data, IMU data and wheel speed sensor data; generating vehicle state data containing vehicle positioning data and attitude data according to the vehicle-end sensor data; Collecting vehicle-end environment perception data, and generating a vehicle-end obstacle data set through a fusion algorithm according to the vehicle-end environment perception data; Collecting roadside environment perception data through a roadside device, and generating a roadside obstacle data set through a perception fusion algorithm according to the roadside environment perception data; The cloud constructs a global obstacle map according to the vehicle-end obstacle data set, the roadside obstacle data set and the vehicle state data, and extracts obstacle data within a preset range centered on the position of each vehicle to generate a vehicle obstacle map for each vehicle; The cloud calculates the global expected speed of each vehicle based on the global obstacle map and all vehicle state data for multi-vehicle collaborative dynamic control; The vehicle-end calculates the vehicle-end expected speed according to the vehicle-end obstacle data set, the roadside obstacle data set and the vehicle obstacle map issued by the cloud; The vehicle-end generates the final control speed according to the vehicle-end expected speed, the map speed limit and the global expected speed; The calculation of the global expected speed of each vehicle comprises the following steps: Obtaining obstacle data within a preset distance in the driving direction of each vehicle from the global obstacle map, and calculating the safe speed required to avoid collision according to the relative position and relative speed of the obstacle and the vehicle; Identifying intersections and turning positions in the lane, and detecting whether there is a path conflict situation in which multiple vehicles approach the same intersection or curve at the same time; When detecting that multiple vehicles have a path conflict, the vehicle passing priority is determined according to the following rules: Calculating the expected time of each vehicle to reach the conflict point; Determining the priority weight according to the vehicle type, load state and driving direction; Generating a vehicle passing sequence according to the priority weight; For vehicles with low priority, the corresponding deceleration is calculated according to the distance of the corresponding vehicle from the conflict point and the current vehicle speed, and the global expected speed after deceleration is generated; For vehicles with high priority, the safe speed calculated based on obstacle avoidance is taken as the global expected speed; For vehicles without path conflict, the safe speed calculated based on obstacle avoidance is taken as the global expected speed; The global expected speed of each vehicle is sent to the corresponding vehicle through 5G communication.

2. The vehicle speed control method based on vehicle-lane-cloud collaboration in the mine unmanned vehicle according to claim 1, wherein: The vehicle state data containing vehicle positioning data and attitude data is generated according to the vehicle-end sensor data, which comprises the following steps: Performing point cloud registration processing on the laser SLAM data to generate vehicle position and attitude estimation based on laser point cloud; Performing feature extraction and matching on the visual SLAM data to generate vehicle position and attitude estimation based on visual features; Obtaining the three-axis acceleration and three-axis angular velocity of the vehicle according to the IMU data; Obtaining the linear speed of the vehicle according to the wheel speed sensor data; Data fusion is performed by using an extended Kalman filter algorithm to obtain vehicle positioning data and attitude data, wherein the positioning data includes three-dimensional position coordinates of the vehicle in the lane coordinate system, and the attitude data includes yaw angle, pitch angle and roll angle.

3. The vehicle-lane-cloud collaborative speed control method for unmanned vehicles in underground mines according to claim 2, characterized in that: obtaining vehicle positioning data and attitude data includes: in the prediction step, based on the vehicle state at the previous time, the position prediction value at the current time is calculated through the kinematics equation by using the three-axis acceleration, and the attitude prediction value at the current time is calculated through the attitude update equation by using the three-axis angular velocity; in the update step, the vehicle position and attitude estimation based on the laser point cloud and the vehicle position and attitude estimation based on the visual features are taken as observations, and the position prediction value and the attitude prediction value are corrected through the Kalman gain; the linear velocity obtained by the wheel speed sensor is compared with the velocity obtained by integrating the acceleration, and when the difference between the two exceeds a preset threshold, the velocity obtained by integrating the acceleration is corrected by using the linear velocity to prevent IMU integral drift; the corrected three-dimensional position coordinates and attitude angles of the vehicle are output as the vehicle positioning data and attitude data.

4. The vehicle-lane-cloud collaborative speed control method for unmanned vehicles in underground mines according to claim 2, characterized in that: generating vehicle-end obstacle data set through fusion algorithm according to vehicle-end environment perception data includes: collecting vehicle-end environment perception data, which includes laser radar point cloud data, millimeter wave radar data and camera image data; according to the laser radar point cloud data, the static point cloud of the mine roadway roof and sidewall is filtered out to obtain dynamic point cloud, and the three-dimensional contour data of the obstacle is obtained by clustering the dynamic point cloud; according to the millimeter wave radar data, the radial velocity and distance of the obstacle detected by the millimeter wave radar are extracted, and the speed of the obstacle relative to the vehicle is calculated according to the Doppler effect; according to the camera image data, the obstacle category is identified through the target detection algorithm; the three-dimensional contour data, the relative speed of the obstacle and the obstacle category are time-aligned and space-registered through the space-time synchronization algorithm, and the multi-sensor fusion algorithm is used to generate the vehicle-end obstacle data set; the vehicle-end obstacle data set is associated with the vehicle state data, and is sent to the surrounding vehicles and roadside devices through V2X communication, and is sent to the cloud server through 5G communication.

5. The vehicle-lane-cloud collaborative speed control method for unmanned vehicles in underground mines according to claim 4, characterized in that: generating vehicle-end obstacle data set through multi-sensor fusion algorithm includes: taking the three-dimensional contour data, the relative speed of the obstacle and the obstacle category as input, and estimating the state of the obstacle through extended Kalman filtering algorithm; according to the preset obstacle category library, the identified obstacle is matched to one of the following categories: mine car, roadway worker, fixed equipment, temporary obstacle or other moving object; the distance and azimuth angle of the obstacle measured by the laser radar are converted into three-dimensional coordinates through coordinate conversion, the distance and azimuth angle measured by the millimeter wave radar are converted into three-dimensional coordinates, and the center point position of the obstacle is calculated through weighted average algorithm, wherein the weight coefficient is adjusted according to the laser radar point cloud density; the radial velocity measured by the millimeter wave radar is decomposed into three-dimensional velocity components; According to the position difference and time interval of the same obstacle in two adjacent frames of laser radar data, the speed estimation value of the obstacle is calculated; The three-dimensional velocity components and the speed estimation value are fused by Kalman filtering to obtain the three-dimensional velocity vector of the obstacle; The principal component analysis is performed on the obstacle point cloud obtained by the laser radar, the principal axis direction is extracted as the obstacle orientation, and the length, width and height of the obstacle are calculated by the minimum circumscribed rectangle algorithm; The distance deviation between the detection positions of the laser radar, the millimeter wave radar and the camera detection bounding box center position is calculated, and the confidence of the corresponding obstacle is calculated according to the distance deviation; The category identification, three-dimensional position coordinates, three-dimensional velocity vector, length, width and height data, orientation angle and confidence of each obstacle are combined to obtain the vehicle-end obstacle data set.

6. The mine unmanned vehicle speed control method based on vehicle-lane-cloud collaboration according to any one of claims 2 to 5, characterized in that: generating a vehicle obstacle map for each vehicle, comprising: the cloud receiving vehicle-end obstacle data sets and vehicle state data sent by multiple vehicles through 5G communication, and roadside obstacle data sets sent by multiple roadside devices through 5G communication or wired network; converting the three-dimensional position coordinates in the vehicle-end obstacle data set relative to the vehicle into global coordinates in the lane coordinate system according to the positioning data of the corresponding vehicle; converting the three-dimensional position coordinates in the roadside obstacle data set relative to the roadside device into global coordinates in the lane coordinate system according to the preset position of the roadside device; matching the same obstacle from different data sources: calculating the Euclidean distance of the global coordinates of the obstacles in different data sources, and determining that they are the same obstacle when the Euclidean distance is less than a preset threshold; fusing the matched same obstacle data: obtaining the obstacle category, length, width, height and orientation angle corresponding to the data source with the highest confidence, and performing weighted average on the position coordinates and velocity vectors of multiple data sources; constructing a global obstacle map according to the global coordinates of the fused obstacle data, wherein each obstacle in the global obstacle map includes: obstacle category identification, global three-dimensional position coordinates, global three-dimensional velocity vector, length, width, height, orientation angle and confidence; extracting the obstacle data within a preset range from the vehicle position according to the vehicle positioning data of each vehicle to generate a vehicle obstacle map for the corresponding vehicle; sending each vehicle obstacle map to the corresponding vehicle through 5G communication.

7. The mine unmanned vehicle speed control method based on vehicle-lane-cloud collaboration according to claim 1, characterized in that: calculating the vehicle-end desired speed, comprising: obtaining the vehicle-end obstacle data set generated by the vehicle-end self-perception, the obstacle data set sent by other vehicles received through V2X communication, the roadside obstacle data set sent by roadside devices received through V2X communication, and the vehicle obstacle map issued by the cloud received through 5G communication; converting all obstacle information to the vehicle coordinate system, correlating the data according to the position of the obstacle, and determining whether the obstacles in different data sources are the same target; fusing the obstacle data associated with the same target: When multiple data sources detect the same obstacle, the obstacle category of the data source with the highest confidence is selected; The position and speed of the obstacle are weighted and averaged; Independent obstacles detected by all data sources are retained; According to the fused obstacle data, effective obstacles representing obstacles located within a preset range in the driving direction of the vehicle are screened; The closest effective obstacle on the driving path is selected, the relative distance and relative speed of the effective obstacle are calculated, and the corresponding vehicle end expected speed is determined through a preset query table which presets the safe vehicle speed under different distance and speed combinations in the underground mine environment; For distance and speed combinations not covered by the query table, a linear interpolation method is used to calculate the vehicle end expected speed.

8. The mine unmanned vehicle speed control method based on vehicle-lane-cloud collaboration according to claim 7, characterized in that: generating the final control speed comprises: obtaining the vehicle end expected speed, the map speed limit of the lane section where the current vehicle is located, and the global expected speed sent by the cloud, selecting the minimum value among the three as the preliminary control speed; obtaining the final control speed of the last control period; calculating the difference between the preliminary control speed and the final control speed of the last control period; when the difference is greater than a product of a preset acceleration threshold value and a control period length, the current control vehicle speed , wherein, is the current control vehicle speed; is a final control vehicle speed of a previous control period; is a preliminary control vehicle speed; sign() is a sign function, returning +1 when the value in the parentheses is positive, and -1 when the value in the parentheses is negative; is a maximum allowed acceleration; T is a control period length; When the difference is less than or equal to a product of a preset acceleration threshold and a control period length, the current control vehicle speed is , wherein, a is a smoothing coefficient, and the value range of a is 0 to 1. Output current control vehicle speed As the final control vehicle speed of the current control cycle.

9. A speed control system for unmanned vehicles in underground mines based on vehicle-lane-cloud cooperation, characterized in that, comprises: a vehicle-mounted intelligent device that collects vehicle end sensor data and generates vehicle state data containing vehicle positioning data and attitude data; collects vehicle end environment perception data and generates vehicle end obstacle data set through fusion algorithm, receives obstacle data sent by other vehicles, roadside devices and cloud, calculates vehicle end expected speed after data fusion, and generates final control speed according to vehicle end expected speed, map speed limit and global expected speed; a roadside intelligent device arranged at turning places, intersections and preset positions of straight lanes in the mine roadway, configured with laser radar, millimeter wave radar and camera; for collecting roadside environment data, generating roadside obstacle data set through perception fusion algorithm, sending roadside obstacle data set to vehicles through V2X communication, and sending roadside obstacle data set to cloud processing device through 5G communication or wired network; an underground communication base station that provides 5G communication link between vehicle-mounted intelligent devices and cloud processing devices, and provides V2X communication link between vehicle-mounted intelligent devices and roadside intelligent devices; a cloud processing device including a data fusion server, a map construction server and a collaborative decision-making server; for receiving vehicle end obstacle data sets of multiple vehicles, vehicle state data and roadside obstacle data sets of multiple roadside devices; constructing a global obstacle map and generating a vehicle obstacle map for each vehicle; based on the global obstacle map and all vehicle state data, calculating the global expected speed of each vehicle for multi-vehicle collaborative control; wherein calculating the global expected speed of each vehicle comprises: obtaining obstacle data within a preset distance in the driving direction of each vehicle from the global obstacle map, and calculating the safe vehicle speed required to avoid collision according to the relative position and relative speed of the obstacle and the vehicle; Identify the intersection and turning position in the tunnel, detect whether there is a path conflict situation of multiple vehicles approaching the same intersection or turning at the same time; When detecting path conflict of multiple vehicles, determine the vehicle passing priority according to the following rules: Calculate the expected time of each vehicle to reach the conflict point; Determine the priority weight according to the vehicle type, load state and driving direction; According to the priority weight, generate the vehicle passing sequence; For vehicles with low priority, calculate the corresponding deceleration according to the distance of the corresponding vehicle from the conflict point and the current vehicle speed, and generate the global expected speed after deceleration; For vehicles with high priority, the safety speed calculated based on obstacle avoidance is taken as the global expected speed; For vehicles without path conflict, the safety speed calculated based on obstacle avoidance is taken as the global expected speed; Send the global expected speed of each vehicle to the corresponding vehicle through 5G communication.

Images