An unmanned driving control system

Abstract

This invention discloses an unmanned driving control system, relating to the field of unmanned driving technology. The system comprises the following components: an environmental perception layer, a localization and modeling layer, a path planning layer, a decision control layer, and a bottom-level execution layer. This invention employs an improved YOLOv7-tiny target detection algorithm and a DeepSort target tracking algorithm, combined with LiDAR point cloud processing, to achieve accurate identification and tracking of track markers such as cones. This not only improves the real-time performance and accuracy of target detection but also enhances the robustness of environmental perception through data fusion technology. Simultaneously, at the path planning and decision control levels, graph search trees and B-spline curves are used to optimize the path, combined with the Stanley / MPC hybrid lateral control algorithm, enabling the race car to maintain a stable trajectory and speed even at high speeds and in complex track environments, significantly improving the stability and accuracy of the unmanned driving system.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, specifically to an autonomous driving control system. Background Technology

[0002] With the rapid development of autonomous driving technology, the performance of autonomous driving control systems, as a core component, directly affects the safety, efficiency, and intelligence level of autonomous vehicles. Especially in high-level competitions such as the Formula Student China competition, the requirements for autonomous driving systems are even more stringent. They not only need to have high-precision environmental perception capabilities and fast and accurate path planning capabilities, but also need to be able to cope with complex and ever-changing competition environments and achieve stable and reliable vehicle control.

[0003] In traditional autonomous racing car designs, the environmental perception layer mainly acquires information about the surrounding environment through sensors such as LiDAR and cameras. However, this approach suffers from problems such as low data fusion efficiency, poor real-time target detection and tracking, and susceptibility to environmental interference. While LiDAR offers high precision, it requires a large amount of data processing and is susceptible to adverse weather conditions such as rain and fog. Although cameras offer a wide field of view, their detection performance is easily affected by significant changes in lighting or the presence of obstructions. Furthermore, traditional path planning methods are mostly based on static maps or simple dynamic obstacle avoidance strategies, making it difficult to cope with complex and ever-changing track environments. Path optimization algorithms also have shortcomings in terms of real-time performance, accuracy, and robustness. These issues limit the effectiveness of autonomous driving technology in complex scenarios.

[0004] To address the aforementioned issues, it is necessary to optimize existing autonomous driving control systems. By introducing an improved YOLOv7-tiny target detection algorithm and a DeepSort target tracking module, combined with LiDAR point cloud processing and fusion algorithms, the real-time performance and accuracy of environmental perception can be enhanced. Simultaneously, a LiDAR SLAM algorithm is employed for real-time map construction and localization, and Delaunay triangulation and graph search tree algorithms are used for path planning to ensure the smoothness and rationality of the path. Therefore, developing an autonomous driving control system that can comprehensively achieve the above characteristics is of great significance. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an unmanned driving control system. This system significantly improves the real-time performance and accuracy of environmental perception by introducing an improved YOLOv7-tiny target detection algorithm and a DeepSort target tracking module, combined with LiDAR point cloud processing and fusion algorithms. Simultaneously, it employs a LiDAR SLAM algorithm for real-time map construction and localization, and combines Delaunay triangulation and graph search tree algorithms for path planning, ensuring the smoothness and rationality of the path. Furthermore, at the decision control layer, it achieves precise control of vehicle motion through a hybrid lateral control algorithm and a segmented speed planning method. In addition, by building a perception experimental platform and adopting an advanced underlying control system, the overall performance and reliability of the system are further improved.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an unmanned driving control system, which includes the following components: an environmental perception layer, a localization and modeling layer, a path planning layer, a decision control layer, and a bottom execution layer;

[0007] The environmental perception layer includes an image processing module, a point cloud processing module, and a data fusion module, and is equipped with a high-precision lidar and a camera. Through the improved YOLOv7-tiny target detection algorithm and DeepSort target tracking algorithm, it can realize real-time detection and tracking of targets such as cones.

[0008] The positioning and mapping layer adopts the SC-LeGO-LOAM laser SLAM algorithm, which constructs a real-time map and performs precise positioning using the scanning data of the LiDAR.

[0009] The path planning layer uses Delaunay triangulation to discretize the global map, generates the target path through the graph search tree algorithm, and uses cubic B-spline curves for path fitting.

[0010] The decision control layer uses a hybrid lateral control algorithm based on DDPG and a three-point curvature velocity planning method to control the front wheel steering angle and acceleration / deceleration of the vehicle based on path information and vehicle status.

[0011] The underlying execution layer receives control signals through the CAN network to precisely control the steering system, braking system, and drive system.

[0012] Furthermore, the environment perception layer includes an image processing module, a point cloud processing module, and a data fusion module. The image processing module obtains the 2D detection box of the target by realizing target detection and tracking. The point cloud processing module realizes target clustering and extracts the target point cloud and 3D coordinate information. The data fusion module analyzes the coordinates, color, and confidence information of the cones in the current field of view by fusing visual and LiDAR data, and provides them to the path planning layer.

[0013] Furthermore, the image processing module in the environment perception layer employs an improved YOLOv7-tiny object detection algorithm to achieve object detection and tracking. Specifically, it receives image data containing target objects and processes the image data using a model. This model includes a feature extraction layer, an object classification layer, and a bounding box regression layer. In the object classification layer, the classification loss L is calculated using the cross-entropy loss function. cls In the bounding box regression layer, the bounding box regression loss L is calculated using the squared error loss function. bbox To optimize the model's ability to predict target bounding boxes, the classification loss and bounding box regression loss are weighted and summed to obtain the total loss, which is calculated using the following formula: L total =αL cls +βL bbox Here, α and β are hyperparameters, and the model is trained and optimized based on the total loss. Based on the ground truth boxes and predicted boxes, the DeepSort object tracking algorithm is used to calculate the CIoU value between each detection box and each predicted box, and an allocation cost matrix is ​​constructed. The formula for calculating CIoU is: Where IoU is the crossover-union ratio, b and b gt ρ represents the center point of the predicted bounding box and the ground truth bounding box, respectively. 2 (b,b gt ) represents the square of the Euclidean distance between the center points, c represents the diagonal distance of the minimum closure region, v represents the aspect ratio consistency measure, and α represents the trade-off parameter. The Hungarian algorithm is used, and based on the above allocation cost matrix, the detection boxes and prediction boxes are matched. A cascade matching mechanism is introduced to improve the matching reliability. Finally, the Kalman filter is updated with the successfully associated detection boxes, and the updated bounding boxes are passed to the next frame as tracking boxes. By combining the recognition results with the target tracking results, the recognition of the cone color can be visualized based on the tracking count. At the same time, the position of the cone target can still be predicted even if the target detector fails to detect the target.

[0014] Furthermore, the point cloud processing module in the environmental perception layer filters out point clouds outside the rectangular areas where cones are distributed, based on the field conditions. It then uses the RANSAC algorithm to select the point cloud with the most randomly fitted points during the iteration process as the extracted ground point cloud, filtering out ground noise points. Radius filtering is then used to remove outliers, resulting in a filtered point cloud. Euclidean clustering is then used to obtain clustered point cloud clusters based on the set cluster sphere radius and clustering point cloud threshold. The height and width of the clustered point cloud clusters are restricted based on the cone characteristics to filter out obstacles outside the cones. Cones that meet the characteristic information are selected, and the extracted cone point clouds and their location information are sent to the data fusion module.

[0015] Furthermore, the environmental perception layer data fusion module receives the cone point cloud clusters and their 3D bounding boxes obtained by the point cloud clustering module, as well as the image bounding boxes obtained by the image processing module. It projects the extreme values ​​of the 3D bounding boxes onto the pixel plane to obtain 2D bounding boxes. Then, it uses the Hungarian algorithm to perform matching calculations and uses the intersection-over-union ratio (IoU) of the distance between the 3D bounding boxes and the 2D bounding boxes as the allocation cost matrix to complete the matching of the image and point cloud detection results, thereby realizing the fusion and association of cone position information and category information.

[0016] Furthermore, the localization and mapping layer adopts the SC-LeGO-LOAM laser SLAM algorithm. It constructs a real-time map using laser radar scanning data and performs precise localization. Specifically, the laser radar scans the surrounding environment and generates point cloud data, extracts feature points from the point cloud data, constructs a real-time map using a feature matching algorithm, and then uses the map and current laser scanning data to perform precise localization using the laser SLAM algorithm. Let the laser radar scanning data be S = {p1, p2, ..., p...} N}, where p i It is a point in the scan, with polar coordinates (r i ,θ i Divide θ into M equal-width intervals, each interval having a width of . For each interval j, calculate the maximum and minimum radii r of the points falling within that interval. max,j r min,j The Scan Context algorithm is used to process the scan data. The Scan Context histogram H is an M-dimensional vector, where H[j] = (r min,j ,r max,j Loop closures are identified by comparing the similarity between the current scan and historical scans. Let the two ScanContext histograms be H1 and H2, then their similarity D can be obtained by calculating the Euclidean distance between corresponding intervals of the two histograms.

[0017] Furthermore, the path planning layer uses the Delaunay triangulation algorithm to discretize the global map into a grid composed of points and lines. Within this discretized grid, a graph search tree algorithm is employed to generate the optimal path from the starting point to the ending point based on the actual and estimated costs between nodes. The generated path points are then used as control points, and a cubic B-spline curve is used for fitting to obtain a smooth path with continuous curvature. The expression for the cubic B-spline curve is: Where P(t) is the position of any point on the path, B j,3 (t) is a cubic B-spline basis function, P i+j It is a control point.

[0018] Furthermore, the decision control layer receives path information generated by the path planning layer and the vehicle's current state information. Based on the vehicle's current speed, it employs a hybrid lateral control algorithm based on DDPG. In the low-speed range, the Stanley algorithm is used to calculate the front wheel angle, while in the high-speed range, the MPC algorithm is used. In the transition range, the weights of the Stanley and MPC algorithms are dynamically adjusted using the DDPG algorithm to achieve a smooth transition. By adopting a three-point curvature velocity planning method, the vehicle's speed is calculated and planned in segments according to the curvature changes of the path, enabling the vehicle to accelerate and decelerate smoothly during driving. The calculated front wheel angle and speed are used as control commands and sent to the underlying actuators through the vehicle control system to control the vehicle's front wheel angle and acceleration / deceleration.

[0019] Furthermore, the underlying execution layer employs a PID control algorithm. Based on the error between the desired steering angle and the actual steering angle and its rate of change, it calculates the output steering angle and controls the steering actuator by adjusting the proportional, integral, and derivative gains. It also employs a fuzzy PID control algorithm, taking the actual speed, desired speed, speed error, and its rate of change as inputs, and dynamically adjusts the PID parameters through fuzzy inference to achieve precise control of the vehicle braking system.

[0020] Compared with existing technologies, this autonomous driving control system has the following advantages:

[0021] I. This invention employs an improved YOLOv7-tiny target detection algorithm and DeepSort target tracking algorithm, combined with LiDAR point cloud processing, to achieve accurate identification and tracking of track markers such as cones. This not only improves the real-time performance and accuracy of target detection but also enhances the robustness of environmental perception through data fusion technology. Furthermore, at the path planning and decision control levels, it uses graph search trees and B-spline curves to optimize the path, combined with the Stanley / MPC hybrid lateral control algorithm, enabling the race car to maintain a stable driving trajectory and speed even at high speeds and in complex track environments, significantly improving the stability and accuracy of the autonomous driving system.

[0022] Second, this invention introduces laser SLAM technology and uses the SC-LeGO-LOAM algorithm to construct a real-time track environment map and perform precise positioning, providing reliable navigation information for the race car. This not only improves the autonomous navigation capability of the race car in unknown or complex environments, but also enhances the system's adaptability and robustness in the face of track changes and emergencies. Furthermore, by introducing the DDPG reinforcement learning method, the weight switching optimization of the Stanley and MPC control algorithms is achieved, realizing a smooth transition of control strategies at different speed ranges, further improving the intelligence level and overall performance of the autonomous driving system.

[0023] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0025] Figure 1 This is a schematic diagram of the process of an unmanned driving control system.

[0026] Figure 2 A flowchart of an unmanned driving control system;

[0027] Figure 3 This is a flowchart of point cloud processing for an unmanned driving control system. Detailed Implementation

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0029] Example 1

[0030] This embodiment describes in detail the specific application process of an unmanned driving control system. The system includes an environmental perception layer, a localization and modeling layer, a path planning layer, a decision control layer, and a bottom execution layer. The layers work together through an efficient data transmission and processing mechanism to achieve autonomous navigation and precise control of the race car.

[0031] The environmental perception layer includes sensor and control hardware, an image processing module, a point cloud processing module, and a data fusion module. The main sensor is a 32-line LiDAR from Beike Tianhui, used for point cloud data acquisition of targets such as cones. Simultaneously, a Daheng industrial wide-angle camera is used for real-time image acquisition with a high acquisition frequency and wide viewing angle. A Guangzhou Daoyuan GNSS / IMU integrated navigation system is selected to obtain high-precision vehicle position, heading angle, and speed information. The industrial control computer is a VIOORBRAV-7601-S001 model, possessing powerful computing capabilities and utilizing a dedicated graphics card for GPU acceleration. The vehicle controller is Huahai Technology Rapid. The ECU-U2 supports MATLAB / Simulink / Stateflow / ECUCoder, facilitating rapid prototype development and algorithm verification. The image processing module employs an improved YOLOv7-tiny target detection algorithm, combined with the DeepSort target tracking module, to achieve real-time detection and tracking of cones. The point cloud processing module processes the raw point cloud using methods such as pass-through filtering, RANSAC algorithm, radius filtering, and Euclidean clustering to extract cone point cloud information. The data fusion module fuses image and point cloud data through joint calibration and time synchronization to obtain precise location and category information of the cones.

[0032] The localization and mapping process employs the SC-LeGO-LOAM laser SLAM algorithm. It constructs a real-time map using LiDAR scan data and performs precise localization. Specifically, the LiDAR scans the surrounding environment and generates point cloud data. Feature points are extracted from the point cloud data, and a real-time map is constructed using a feature matching algorithm. Using the map and the current LiDAR scan data, precise localization is achieved through the laser SLAM algorithm. Let the LiDAR scan data be S = {p1, p2, ..., p...}. N}, where p i It is a point in the scan, with polar coordinates (r i ,θ i Divide θ into M equal-width intervals, each interval having a width of . For each interval j, calculate the maximum and minimum radii r of the points falling within that interval. max,j r min,j The Scan Context algorithm is used to process the scan data. The Scan Context histogram H is an M-dimensional vector, where H[j] = (r min,j ,r max,j Loop closures are identified by comparing the similarity between the current scan and historical scans. Let the two Scan Context histograms be H1 and H2, then the similarity D between them can be obtained by calculating the Euclidean distance between corresponding intervals of the two histograms. In practical applications, by continuously scanning with LiDAR to acquire real-time point cloud data, and processing and analyzing this data, the extracted feature points and the constructed real-time map are combined with the current LiDAR scan data to achieve precise positioning through the LiDAR SLAM algorithm. At the same time, by identifying loop closures, the map can be continuously optimized and updated to improve its accuracy and completeness. In this way, the autonomous driving system can more accurately perceive the surrounding environment, achieve precise positioning and navigation, and ensure driving safety and stability.

[0033] The path planning layer uses the Delaunay triangulation algorithm to discretize the global map into a grid composed of points and lines. The Delaunay triangulation grid is generated between cones using an incremental algorithm. The advantage of the incremental algorithm is that it can dynamically add identified cones as grid nodes, allowing the grid to be updated as the race car moves. Triangulation discretizes the global map, resulting in a global map containing only points and lines, which is a typical graph search problem. Therefore, the graph search tree algorithm is chosen for path generation. First, nodes are sampled in the Delaunay triangulation grid space. Filters are used to remove severely non-compliant nodes according to rules. Then, using an N-ary decision tree method, the optimal path from the starting point to the ending point is generated based on the actual cost and estimated cost between nodes. The generated path points are used as control points, and a cubic B-spline curve is used for fitting to obtain a smooth path with continuous curvature. The expression for the cubic B-spline curve is: Where P(t) is the position of any point on the path, B j,3 (t) is a cubic B-spline basis function, P i+j It is a control point.

[0034] The decision control layer receives path information generated by the path planning layer and the vehicle's current state information. Based on the vehicle's current speed, it employs a hybrid lateral control algorithm based on DDPG. In the low-speed range, the Stanley algorithm is used to calculate the front wheel steering angle, and its calculation formula is as follows: Among them, K p and K d These are the proportional and differential gains, respectively. θ is the deviation angle between the vehicle's current heading and the target path. The MPC algorithm is used to calculate the front wheel steering angle in the high-speed range, and the DDPG algorithm dynamically adjusts the weights of the Stanley and MPC algorithms in the transition range to achieve a smooth transition. The calculation formula is as follows: The constraint is: stx k+1 =f(x) k ,u k ),x0=x initial ,x k ∈X,u k∈U, where N is the prediction time domain, Q and R are weight matrices, f is the vehicle dynamics model, and X and U are the constraint sets for the state and control inputs, respectively. k and x ref,k These represent the state parameters at different path points. By employing the three-point curvature velocity planning method, the vehicle speed is calculated and planned in segments based on the curvature changes of the path, enabling the vehicle to accelerate and decelerate smoothly during driving. The calculated front wheel angle and speed are used as control commands and sent to the underlying actuators through the vehicle control system to achieve control over the vehicle's front wheel angle and acceleration / deceleration.

[0035] The underlying execution layer employs a PID control algorithm. Based on the error between the desired and actual steering angles and its rate of change, it calculates the output steering angle and controls the steering actuator by adjusting the proportional, integral, and derivative gains. Where e(t) is the error between the actual rotation angle and the expected rotation angle, and K p K i K d These are proportional, integral, and derivative gains, respectively. A fuzzy PID control algorithm is employed, with actual speed, desired speed, speed error, and its rate of change as inputs. The calculation formula is as follows: Where ΔK p ,ΔK i ,ΔK d These are the adjustment values ​​for proportional, integral, and derivative gains, respectively. By dynamically adjusting the PID parameters through fuzzy inference, precise control of the vehicle braking system can be achieved.

[0036] In summary, this embodiment achieves high-precision and high-stability autonomous driving in complex environments by integrating advanced sensor technology, optimized algorithms, and precise control strategies.

[0037] Example 2

[0038] Based on Embodiment 1, this embodiment describes in detail the specific application of the environmental perception layer in an unmanned driving control system.

[0039] The environment perception layer includes an image processing module, a point cloud processing module, and a data fusion module. The image processing module performs target detection and tracking to obtain the 2D detection box of the target. The point cloud processing module performs target clustering and extracts the target point cloud and 3D coordinate information. The data fusion module merges visual and LiDAR data to parse the coordinates, color, and confidence information of the cones in the current field of view, which are then provided to the path planning layer.

[0040] The sensor uses Beike Tianhui 32-line LiDAR as the main sensor to collect point cloud data of targets such as cones. At the same time, it is equipped with Daheng Industrial Wide-Angle Camera to collect real-time images with a high acquisition frequency and viewing angle. Guangzhou Daoyuan GNSS / IMU integrated navigation is selected to obtain high-precision vehicle position, heading angle and speed information. The industrial control computer is VIOORBRAV-7601-S001 model, which has powerful computing capabilities and uses a dedicated graphics card for GPU acceleration computing. The vehicle controller is Huahai Technology RapidECU-U2, which supports MATLAB / Simulink / Stateflow / ECUCoder, which facilitates rapid prototype development and algorithm verification.

[0041] The image processing module employs an improved YOLOv7-tiny object detection algorithm for object detection and tracking. Specifically, it receives image data containing target objects and processes the image data using a model. This model includes a feature extraction layer, an object classification layer, and a bounding box regression layer. In the object classification layer, the classification loss is calculated using the cross-entropy loss function, and its calculation formula is as follows: Where, p i It represents the probability of each type, y. i It is image information, L cls This is the classification loss, where C represents the number of categories. In the bounding box regression layer, the bounding box regression loss L is calculated using the squared error loss function. bbox To optimize the model's ability to predict target bounding boxes, the calculation formula is as follows: Among them, L bbox It is the bounding box regression loss. It is the actual bounding box, t i It is the predicted bounding box, λ i These are the weights of different parameters, where x and y are the center coordinate offsets, and w and h are the width and height scaling. The classification loss and bounding box regression loss are weighted and summed to obtain the total loss, which is calculated using the formula: L total =αL cls +βL bbox Here, α and β are hyperparameters, and the model is trained and optimized based on the total loss. Based on the ground truth boxes and predicted boxes, the DeepSort object tracking algorithm is used to calculate the CIoU value between each detection box and each predicted box, and an allocation cost matrix is ​​constructed. The formula for calculating CIoU is: Where IoU is the crossover-union ratio, b and b gt ρ represents the center point of the predicted bounding box and the ground truth bounding box, respectively. 2 (b,b gt) represents the square of the Euclidean distance between the center points, c represents the diagonal distance of the minimum closure region, v represents the aspect ratio consistency measure, and α represents the trade-off parameter. The Hungarian algorithm is used, and based on the above allocation cost matrix, the detection boxes and prediction boxes are matched. A cascade matching mechanism is introduced to improve the matching reliability. Finally, the Kalman filter is updated with the successfully associated detection boxes, and the updated bounding boxes are passed to the next frame as tracking boxes. By combining the recognition results with the target tracking results, the recognition of the cone color can be visualized based on the tracking count. At the same time, the position of the cone target can still be predicted even if the target detector fails to detect the target.

[0042] In the perception layer, the point cloud processing module filters out point clouds outside the rectangular areas of the cone distribution based on the field conditions. Then, using the RANSAC algorithm, it selects the point cloud with the most randomly fitted points during the iteration process as the extracted ground point cloud, based on a set threshold. Specifically, three points are randomly selected from the point cloud data, and the plane equation is calculated using these three points. The plane equation can be expressed as: Ax + By + Cz + D = 0, where A, B, and C are the normal vectors of the plane, and D is the distance from the plane to the origin. For each point in the point cloud, its distance to the plane is calculated. If the distance is less than a certain threshold t, the point is considered an interior point. The plane equations are recalculated for interior points to improve model accuracy. This process is repeated multiple times, with the number of interior points calculated in each iteration. The iteration with the most interior points is selected as the final result. Ground noise points in the final result are filtered out, and outliers are removed using radius filtering to obtain the filtered point cloud. Euclidean clustering is then used to obtain clustered point cloud clusters based on the set cluster sphere radius and cluster point cloud threshold. The height and width of the clustered point cloud clusters are restricted based on the cone feature to filter out obstacles outside the cones. Cones that meet the feature information are selected, and the extracted cone point clouds and their location information are sent to the data fusion module.

[0043] The data fusion module receives the cone point cloud clusters and their 3D bounding boxes obtained by the point cloud clustering module, as well as the image bounding boxes obtained by the image processing module. Then, it projects the extreme values ​​of the 3D bounding boxes onto the pixel plane to obtain 2D bounding boxes. Next, it uses the Hungarian algorithm to perform matching calculations and uses the distance intersection-union ratio of the points to the point cloud as the allocation cost matrix to complete the matching of the image and point cloud detection results, thereby realizing the fusion and association of cone position information and category information.

[0044] In summary, this embodiment, by employing the improved YOLOv7-tiny target detection algorithm and the DeepSort target tracking module, combined with LiDAR point cloud processing, achieves accurate identification and tracking of track markers such as cones. This not only improves the real-time performance and accuracy of target detection but also further enhances the robustness of environmental perception through data fusion technology.

[0045] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An unmanned driving control system, characterized in that, The system comprises the following components: environmental perception layer, localization and modeling layer, path planning layer, decision control layer, and underlying execution layer; The environmental perception layer includes an image processing module, a point cloud processing module, and a data fusion module, and is equipped with a high-precision lidar and a camera. Through the improved YOLOv7-tiny target detection algorithm and DeepSort target tracking algorithm, it can realize real-time detection and tracking of cone targets. The positioning and mapping layer adopts the SC-LeGO-LOAM laser SLAM algorithm, which constructs a real-time map and performs precise positioning using the scanning data of the LiDAR. The path planning layer uses Delaunay triangulation to discretize the global map, generates the target path through the graph search tree algorithm, and uses cubic B-spline curves for path fitting. The decision control layer uses a hybrid lateral control algorithm based on DDPG and a three-point curvature velocity planning method to control the front wheel steering angle and acceleration / deceleration of the vehicle based on path information and vehicle status. The underlying execution layer receives control signals through the CAN network to precisely control the steering system, braking system, and drive system.

2. The unmanned driving control system according to claim 1, characterized in that, The environment perception layer includes an image processing module, a point cloud processing module, and a data fusion module. The image processing module performs target detection and tracking to obtain a 2D bounding box for the target. The point cloud processing module performs target clustering and extracts the target point cloud and 3D coordinate information. The data fusion module fuses visual and LiDAR data to parse the coordinates, color, and confidence information of the cones in the current field of view, which are then provided to the path planning layer.

3. The unmanned driving control system according to claim 2, characterized in that, The image processing module in the environment perception layer employs an improved YOLOv7-tiny object detection algorithm for object detection and tracking. Specifically, it receives image data containing target objects and processes the image data using a model. This model includes a feature extraction layer, an object classification layer, and a bounding box regression layer. In the object classification layer, the classification loss L is calculated using the cross-entropy loss function. cls In the bounding box regression layer, the bounding box regression loss L is calculated using the squared error loss function. bbox To optimize the model's ability to predict target bounding boxes, the classification loss and bounding box regression loss are weighted and summed to obtain the total loss, which is calculated using the following formula: L total =αL cls +βL bbox Here, α and β are hyperparameters, and the model is trained and optimized based on the total loss. Based on the ground truth boxes and predicted boxes, the DeepSort object tracking algorithm is used to calculate the CIoU value between each detection box and each predicted box, and an allocation cost matrix is ​​constructed. The formula for calculating CIoU is: Where IoU is the crossover-union ratio, b and b gt ρ represents the center point of the predicted bounding box and the ground truth bounding box, respectively. 2 (b,b gt ) represents the square of the Euclidean distance between the center points, c represents the diagonal distance of the minimum closure region, v represents the aspect ratio consistency measure, and α represents the trade-off parameter. The Hungarian algorithm is used, and based on the above allocation cost matrix, the detection boxes and prediction boxes are matched. A cascade matching mechanism is introduced to improve the matching reliability. Finally, the Kalman filter is updated with the successfully associated detection boxes, and the updated bounding boxes are passed to the next frame as tracking boxes. By combining the recognition results with the target tracking results, the recognition of the cone color can be visualized based on the tracking count. At the same time, the position of the cone target can still be predicted even if the target detector fails to detect the target.

4. The unmanned driving control system according to claim 2, characterized in that, The point cloud processing module in the environmental perception layer filters out point clouds outside the rectangular areas where cones are distributed, based on the field conditions. It then uses the RANSAC algorithm to select the point cloud with the most randomly fitted points during the iteration process as the extracted ground point cloud, filtering out ground noise points. Radius filtering is then used to remove outliers, resulting in a filtered point cloud. Euclidean clustering is then used to obtain clustered point cloud clusters based on the set cluster sphere radius and clustering point cloud threshold. The height and width of the clustered point cloud clusters are restricted based on the cone characteristics to filter out obstacles outside the cones. Cones that meet the characteristic information are selected, and the extracted cone point clouds and their location information are sent to the data fusion module.

5. The unmanned driving control system according to claim 2, characterized in that, The environmental perception layer data fusion module receives the cone point cloud clusters and their 3D bounding boxes obtained by the point cloud clustering module, as well as the image bounding boxes obtained by the image processing module. It projects the extreme values ​​of the 3D bounding boxes onto the pixel plane to obtain 2D bounding boxes. Then, it uses the Hungarian algorithm to perform matching calculations and uses the intersection-over-union ratio (IoU) of the distance between the 3D bounding boxes and the 2D bounding boxes as the allocation cost matrix to complete the matching of the image and point cloud detection results, thereby realizing the fusion and association of cone position information and category information.

6. The unmanned driving control system according to claim 1, characterized in that, The localization and mapping layer construction adopts the SC-LeGO-LOAM laser SLAM algorithm. A real-time map is built using LiDAR scanning data for precise localization. Specifically, the LiDAR scans the surrounding environment and generates point cloud data. Feature points are extracted from the point cloud data, and a real-time map is constructed using a feature matching algorithm. Using the map and the current laser scanning data, precise localization is achieved through the laser SLAM algorithm. Let the LiDAR scanning data be S = {p1, p2, ..., pN}, where pi is a point in the scan with polar coordinates (ri, θi). θ is divided into M equal-width intervals, each with a width of... For each interval j, calculate the maximum and minimum radii r of the points falling within that interval. max,j r min,j The Scan Context algorithm is used to process the scanned data. The Scan Context histogram H is an M-dimensional vector, where H[j] = (r min,j ,r max,j Loop closures are identified by comparing the similarity between the current scan and historical scans. Let the two Scan Context histograms be H1 and H2, then the similarity D between them can be obtained by calculating the Euclidean distance between corresponding intervals of the two histograms.

7. The unmanned driving control system according to claim 1, characterized in that, The path planning layer uses the Delaunay triangulation algorithm to discretize the global map into a grid composed of points and lines. Within this discretized grid, a graph search tree algorithm is employed to generate the optimal path from the starting point to the ending point based on the actual and estimated costs between nodes. The generated path points are used as control points, and a cubic B-spline curve is used for fitting to obtain a smooth path with continuous curvature. The expression for the cubic B-spline curve is: Where P(t) is the position of any point on the path, B j,3 (t) is a cubic B-spline basis function, P i+j It is a control point.

8. The unmanned driving control system according to claim 1, characterized in that, The decision control layer receives path information and vehicle current state information generated by the path planning layer. Based on the vehicle's current speed, it employs a hybrid lateral control algorithm based on DDPG. In the low-speed range, the Stanley algorithm is used to calculate the front wheel angle, while in the high-speed range, the MPC algorithm is used. In the transition range, the weights of the Stanley and MPC algorithms are dynamically adjusted using the DDPG algorithm to achieve a smooth transition. By adopting a three-point curvature velocity planning method, the vehicle speed is calculated and planned in segments according to the curvature changes of the path, enabling the vehicle to accelerate and decelerate smoothly during driving. The calculated front wheel angle and speed are used as control commands and sent to the underlying actuators through the vehicle control system to control the vehicle's front wheel angle and acceleration / deceleration.

9. The unmanned driving control system according to claim 1, characterized in that, The underlying execution layer employs a PID control algorithm. Based on the error between the desired steering angle and the actual steering angle and its rate of change, it calculates the output steering angle and controls the steering actuator by adjusting the proportional, integral, and derivative gains. It also employs a fuzzy PID control algorithm, taking the actual speed, desired speed, speed error, and its rate of change as inputs. Through fuzzy inference, it dynamically adjusts the PID parameters to achieve precise control of the vehicle braking system.

Images