Based on the autonomous object detection and obstacle avoidance intelligent recognition system of unmanned vehicles

Abstract

This invention discloses an intelligent obstacle avoidance system based on autonomous object detection at the orientation end of an unmanned vehicle, comprising the following modules: a perception data acquisition module for collecting and preprocessing multi-source perception data to generate an environmental perception dataset; a feature extraction module for inputting the environmental perception dataset into a Motion-DETR dynamic perception model to generate a target object feature set; a perception map construction module for constructing a dynamic environmental perception map; a threat assessment module for calculating threat levels, generating an orientation-end threat level matrix, and establishing a dynamic risk distribution map; a path expansion optimization module for dynamically selecting the expansion direction using a path expansion optimization method that incorporates dynamic risk costs; a path cost optimization module for dynamically optimizing the expansion structure based on the principle of minimizing cumulative path cost; and a path output module for generating the optimal safe driving path in a dynamic environment. This invention provides an intelligent obstacle avoidance path planning method that integrates perception mapping and risk optimization.

Description

Technical Field

[0001] This invention relates to the fields of intelligent driving and artificial intelligence perception and decision-making technology, and in particular to an intelligent identification system for autonomous object detection and obstacle avoidance based on the orientation of unmanned vehicles. Background Technology

[0002] In the field of intelligent driving technology, how to achieve efficient autonomous detection and safe obstacle avoidance of unmanned vehicles in dynamic and complex environments has always been a core issue in the interdisciplinary research of autonomous driving, robot perception, and intelligent decision-making. With the continuous expansion of unmanned vehicle application scenarios, the demands for real-time environmental perception, accuracy, and robustness of autonomous decision-making are constantly increasing. Traditional obstacle avoidance methods relying on rule-driven or single perception modes are gradually failing to meet the requirements for efficient and safe driving in complex dynamic environments. To improve the environmental understanding and obstacle avoidance decision-making capabilities of unmanned vehicles in complex roads and scenarios, academia and industry have introduced advanced perception and decision-making technologies based on deep learning, dynamic modeling, and intelligent reasoning. In particular, novel neural network models, represented by Transformer, have become a key research direction in intelligent driving environmental perception and path planning in recent years due to their ability to capture spatiotemporal dynamic features in large-scale data.

[0003] Existing autonomous vehicle environmental perception and obstacle avoidance methods generally rely on LiDAR point cloud data, camera visual information, or millimeter-wave radar signals, combined with classic object detection networks (such as Faster R-CNN and YOLO series) to identify objects in static environments, and obstacle avoidance is performed based on path planning algorithms such as A* search, D*Lite, and traditional RRT. Some methods introduce motion prediction modules in the perception stage, such as linear motion estimation based on Kalman filters and short-term trajectory prediction based on trajectory regression networks, which can make a rough prediction of dynamic targets to some extent. However, most of these technologies are limited to the perception of the environmental state at a single time step, lacking fine-grained modeling of the future movement trends and threat changes of target objects. This leads to problems such as obstacle avoidance delays, frequent path conflicts, or planning failures in complex dynamic environments, making it difficult to meet the dual requirements of safety and real-time performance in highly dynamic traffic environments.

[0004] In terms of dynamic environment understanding, existing technologies typically rely on static image feature extraction for object recognition, neglecting the crucial role of object motion state and trend information in risk assessment. Even when short-term trajectory prediction is introduced, it is mostly independent of threat level modeling, failing to organically integrate motion trends with local threat perception and lacking a unified dynamic risk distribution modeling mechanism. This fragmented approach makes the system prone to local path infeasibility, decision lag, and even emergency braking when encountering high-density, irregularly moving target groups, seriously affecting driving safety and stability.

[0005] In path planning optimization, existing methods based on Rapid Random Tree (RRT) or its improved versions (such as RRT* and Informed RRT*) mostly rely on static environment maps for extended searches. The path cost function is only related to the location of spatial obstacles, lacking a real-time introduction of dynamic threat information and a cost adjustment mechanism. This makes it difficult to generate forward-looking safe driving paths in complex dynamic environments. Especially when facing rapidly approaching dynamic obstacles, traditional RRT, lacking a dynamic risk perception mechanism, is prone to getting stuck in local optima or dead zones in its extended direction, leading to a significant decrease in obstacle avoidance performance.

[0006] Regarding the granularity of environmental modeling, existing methods generally employ a globally unified cost map or coarse-grained region partitioning, failing to perform fine-grained dynamic threat modeling based on the autonomous vehicle's own orientation characteristics. In reality, during actual driving, the importance of different orientations (such as directly in front, left front, and right rear) to driving decisions varies significantly. A single global cost field cannot reflect this heterogeneity of local threats, making it difficult to accurately prioritize avoiding key threat directions during the path planning stage, thus reducing obstacle avoidance efficiency and the rationality of driving decisions.

[0007] Regarding perception models, although some studies have introduced Transformer-based target detection networks, these mainly focus on static object detection, and are insufficient for modeling the continuous motion trends of dynamic targets. Existing publicly available technologies have not systematically integrated dynamic perception networks with location-based threat segmentation, local risk modeling, and dynamic path optimization, lacking a complete system that integrates perception and decision-making with a dynamic feedback loop. Furthermore, existing systems generally lack mechanisms for real-time path adjustment based on dynamic environmental perception updates, leading to delayed path replanning and a sharp increase in security risks once environmental changes are drastic.

[0008] Therefore, current technologies lack an intelligent autonomous obstacle avoidance system capable of accurately extracting the movement trends of target objects based on dynamic perception models, combining directional threat level classification, dynamically establishing future movement trajectories and risk distribution models, and embedding dynamic threat prediction results into the path expansion and optimization process in real time. Especially when facing complex multi-target dynamic environments, high-density traffic flow, or irregularly moving targets, existing technologies still have significant shortcomings in terms of spatiotemporal consistency of environmental understanding, real-time dynamic threat response, and robustness of path planning, necessitating the development of new technical solutions. Summary of the Invention

[0009] One objective of this invention is to propose an intelligent obstacle avoidance system for autonomous object detection and avoidance based on the orientation of unmanned vehicles. This invention integrates dynamic environmental perception modeling with a fast random tree optimization method that incorporates dynamic risk costs to construct an intelligent autonomous obstacle avoidance system based on orientation-based dynamic threat prediction. By introducing a Motion-DETR dynamic perception model to extract the motion trend features of each target object in the environment, a dynamic environmental perception map is constructed in conjunction with the current position of the unmanned vehicle, and an orientation-based threat level matrix is ​​generated according to orientation division rules. Furthermore, a dynamic risk distribution model is established through trajectory prediction and local risk modeling. Threat cost correction factors are dynamically introduced during the path planning stage to optimize the path expansion strategy in real time. This system has advantages such as high environmental perception accuracy, fast dynamic threat response, strong obstacle avoidance decision intelligence, and high path planning robustness. It is suitable for intelligent decision-making tasks of autonomous obstacle detection and avoidance and safe driving of unmanned vehicles in complex dynamic environments.

[0010] The autonomous obstacle detection and avoidance intelligent identification system for unmanned vehicles according to an embodiment of the present invention includes the following modules:

[0011] The perception data acquisition module is used to collect multi-source perception data of the environment around the unmanned vehicle, perform time synchronization, spatial alignment and standardization preprocessing, and generate an environmental perception dataset.

[0012] The feature extraction module is used to input the environmental perception dataset into the Motion-DETR dynamic perception model, extract information about each target object in the environment, and generate a target object feature set.

[0013] The perception map construction module is used to construct a dynamic environment perception map based on the feature set of the target object, with the current position of the unmanned vehicle as the center, and divide the dynamic environment perception map into multiple directional end regions according to the preset orientation division rules.

[0014] The threat assessment module is used to calculate the threat level based on the information of the target object in each directional area, generate a directional threat level matrix, and predict the future movement trajectory by combining the movement trend of the target object to establish a dynamic risk distribution map.

[0015] The path expansion and optimization module is used to take the dynamic risk distribution map as input, set the current position of the unmanned vehicle as the starting point of path planning, and adopt a path expansion and optimization method that introduces dynamic risk costs to dynamically select the expansion direction.

[0016] The path cost optimization module is used to update the cumulative path cost of each expansion node in real time during the path expansion process, and dynamically optimize the expansion structure based on the principle of minimizing the cumulative path cost.

[0017] The path output module is used to backtrack from the endpoint node to the starting node along the parent node link to generate the optimal safe driving path in a dynamic environment.

[0018] This system, through a modular design, organically integrates six core functions: environmental perception, dynamic graph modeling, threat assessment, path expansion, path cost optimization, and path output, constructing an intelligent obstacle avoidance system architecture with complete closed-loop capabilities. Clear data flows and functional boundaries exist between modules, ensuring functional coordination and consistent response in complex dynamic scenarios. The modular design not only enhances the system's maintainability and scalability but also provides strong support for subsequent algorithm optimization and system iteration, greatly improving the system's adaptability and engineering capabilities across different platforms and scenarios.

[0019] The intelligent obstacle identification method for autonomous object detection and avoidance based on the orientation of an unmanned vehicle according to an embodiment of the present invention includes the following steps:

[0020] S1. Collect multi-source perception data of the environment around the unmanned vehicle, and perform time synchronization, spatial alignment and standardization preprocessing on the multi-source perception data to generate an environmental perception dataset.

[0021] S2. Input the environmental perception dataset into the Motion-DETR dynamic perception model to extract information about each target object in the environment and form a target object feature set.

[0022] S3. Based on the target object feature set, construct a dynamic environment perception map with the current position of the unmanned vehicle as the center, and divide the dynamic environment perception map into multiple directional end regions according to the preset orientation division rules.

[0023] S4. Within each azimuth region, calculate the local threat score based on the target object information and generate the azimuth threat level matrix.

[0024] S5. Based on the azimuth threat level matrix and the target object's motion trend information, predict the target object's future trajectory and establish a dynamic risk distribution map;

[0025] S6. Based on the dynamic risk distribution map, the current position of the unmanned vehicle is set as the starting point of the path planning. The RRT path optimization method with dynamic risk cost is used to expand the path search. The future trajectory and threat level of the target object are used as path cost correction factors to update the dynamic risk cost path expansion tree.

[0026] S7. During the path expansion search process, the cumulative cost of each node is updated in real time, and the expansion direction is dynamically selected based on the principle of minimum cost to generate the optimal safe driving path in the dynamic environment.

[0027] By constructing a comprehensive processing framework encompassing multi-source perception data acquisition, feature extraction, environmental mapping, risk modeling, and path planning and output, this invention introduces the Motion-DETR dynamic perception model and a risk distribution modeling method based on threat level and movement trends to achieve efficient linkage between environmental perception and path decision-making. This solution can simultaneously identify environmental changes and potential risk factors in dynamic multi-objective environments, and deeply integrate risk control and path feasibility calculation through the dynamic generation and real-time optimization of path expansion trees. Compared to traditional path planning methods that rely on static maps or single-objective modeling, this invention significantly improves the safety, consistency, and environmental adaptability of path planning under complex conditions such as multi-source interference, motion conflicts, and spatial changes, meeting the key requirements of highly reliable obstacle avoidance and autonomous navigation for unmanned vehicles.

[0028] Optionally, S1 specifically includes:

[0029] S11. Collect multi-source perception data of the environment around the unmanned vehicle, including visual image data and three-dimensional point cloud data.

[0030] S12. Perform time synchronization processing on multi-source sensing data and extract the acquisition timestamp t of the visual image data. c The timestamp of the 3D point cloud data acquisition r Using the nearest neighbor matching strategy, select t c and t r Pair visual image data and 3D point cloud data with the smallest absolute time difference;

[0031] S13. Assign a unified synchronization timestamp to the paired visual image data and 3D point cloud data. The unified synchronization timestamp is set to the acquisition timestamp t of the visual image data. c The timestamp of the 3D point cloud data acquisition r The average value;

[0032] S14. Perform spatial alignment processing on the time-synchronized multi-source sensing data, based on the extrinsic parameter matrix T. ext The intrinsic parameter matrix K transforms the 3D point cloud data in the radar coordinate system to the camera coordinate system, and projects the transformed 3D point cloud coordinates onto the image plane to obtain the corresponding 2D pixel positions. Points that exceed the boundary of the image plane after projection are removed, thus realizing the fusion of visual image data and 3D point cloud data in a unified spatial reference system.

[0033] S15. Standardize the multi-source perception data after time synchronization and spatial alignment. The visual image data is normalized to standardize the pixel values ​​to the range [0,1]. The three-dimensional point cloud data is standardized according to the set ranging range to standardize the ranging values ​​to the range [0,1].

[0034] S16. The standardized multi-source sensing data is fused and organized according to a unified spatial indexing rule to generate an environmental sensing dataset.

[0035] By performing time synchronization, spatial alignment, and standardized preprocessing on the raw sensing data, the problems of inconsistent sampling frequencies, inconsistent coordinate systems, and large scale fluctuations in multi-source sensor data were resolved, ensuring the accuracy and consistency of the sensing data from the source. In particular, the use of spatial affine mapping and linear interpolation for alignment and filling effectively improved the accuracy of describing the continuous motion state of dynamic objects, providing a high-quality input foundation for subsequent target detection and graph modeling processes, thereby significantly improving the stability and reliability of the overall system performance.

[0036] Optionally, S2 specifically includes:

[0037] S21. Input the environmental perception dataset into the Motion-DETR dynamic perception model;

[0038] S22. Using a convolutional neural network set within the visual feature extractor module, the visual image data in the environmental perception dataset is encoded to extract visual spatial feature information and generate a visual feature tensor F. img ;

[0039] S23. Using a multilayer perceptron network set within the point cloud feature extractor module, the 3D point cloud data in the environmental perception dataset is encoded point-by-point to extract the spatial geometric features of the 3D point cloud and generate the point cloud feature tensor F. pc ;

[0040] S24. For the visual feature tensor F respectively... img and point cloud feature tensor F pc Introducing a self-attention weight matrix for feature reweighting:

[0041]

[0042] Among them, F′ img For the weighted visual feature tensor, F′ pc For the weighted point cloud feature tensor, Let be the self-attention weighting matrix corresponding to the visual feature tensor. Let be the self-attention weighting matrix corresponding to the point cloud feature tensor;

[0043] S25. Weighted visual feature tensor F′ img With weighted point cloud feature tensor F′ pc Perform additive fusion processing to generate a fused feature tensor through the fusion module:

[0044] F fusion =α·F′ img +(1-α)·F′ pc ;

[0045] Among them, F fusion Let α be the fusion feature tensor, and α be the fusion weight coefficient.

[0046] S26, Fen the feature tensor F fusion The input is fed into the dynamic feature modeling module, where a multi-head self-attention mechanism is used to model and fuse the dynamic relationships between feature units, outputting a target object feature set O. This target object feature set includes category information, center position coordinates, velocity vector, and motion trend features.

[0047]

[0048] Among them, o i Let c be the feature vector of the i-th target object. i For the category information of the i-th target object, (x i ,y i () represents the center coordinates of the i-th target object. Let θ be the velocity vector of the i-th target object in the x and y directions. i Let represent the motion trend characteristics of the i-th target object.

[0049] By employing the Motion-DETR dynamic perception model to perform target detection and feature extraction on standardized perception data, the system can simultaneously achieve spatial localization and motion state estimation, outputting a high-dimensional feature set that comprehensively characterizes key elements such as the target object's category, position, velocity, and orientation. This model combines the long-range dependency modeling capabilities of the Transformer structure with the detection efficiency advantages of DETR, improving target recognition accuracy while maintaining real-time performance. It exhibits particularly strong recognition and separation capabilities in complex scenarios such as high-speed moving targets, small targets, and multiple overlapping targets, enhancing the system's adaptability to dynamic environments.

[0050] Optionally, S3 specifically includes:

[0051] S31. Based on the target object feature set O, extract the center position coordinates (x, y, y) of the target object. i ,y i ), velocity vector and motion trend characteristics θi Using the current position coordinates (x0, y0) of the unmanned vehicle as a reference, the predicted position is calculated based on the center position coordinates and velocity vector of the target object and the set short-term prediction step size. Then, the dynamic polar radius is determined based on the Euclidean distance between the predicted position and the current position of the unmanned vehicle, and the dynamic polar angle is determined based on the directional relationship between the position of the target object and the unmanned vehicle after the short-term prediction.

[0052] S32. Based on the dynamic polar radius, select effective target objects that meet the sensing radius condition and construct a subset of effective target objects. ′ The perception radius is the maximum perception distance threshold set with the current position of the unmanned vehicle as the center, which limits the processing object to the physical space area that the unmanned vehicle can perceive and react to in real time.

[0053] S33. Centered on the current position of the unmanned vehicle, based on the effective target object subset O ′ A dynamic environment perception map is constructed, wherein the nodes of the dynamic environment perception map are effective target objects, and the node attributes include the predicted position, velocity vector and motion trend characteristics of the effective target objects.

[0054] S34. Based on the dynamic polar angles corresponding to the nodes of the dynamic environment perception map, and according to the preset direction division rules, divide the dynamic environment perception map into N azimuth end regions, each azimuth end region being numbered A. j , where j∈{1,2,...,N}.

[0055] A dynamic environmental perception map is constructed based on the feature set of target objects, and a polar coordinate-based orientation region modeling mechanism is introduced to improve the spatial resolution of threat information from different directions. This map structure not only preserves the spatial dependencies between various objects in the environment but also achieves independent expression of local orientation risks through a partitioning strategy, making it particularly suitable for structural analysis of densely perceived areas along the autonomous vehicle's forward direction. Compared to traditional planar perception maps, this method improves the response accuracy of orientation-based obstacle avoidance strategies, providing a spatial structural foundation for subsequent threat assessment and path planning.

[0056] Optionally, S4 specifically includes:

[0057] S41. Based on the dynamic environment perception map, for each azimuth end region A j Extract region nodes and construct a subset O of region nodes. j Among them, the region node subset O j Each region node o i This includes predicted position, velocity vector, motion trend characteristics, and dynamic polar radius;

[0058] S42, For each region node o i Stored velocity vector calculation region node oi velocity modulus s i ;

[0059] S43, Based on each region node o i Current motion trend characteristics θ i The motion trend characteristics θ of the previous moment i (t-1) Calculate the rate of change of motion trend Δθ i ;

[0060] S44, Statistical region node o i The number of neighboring nodes within the perception radius, n i As a local density feature;

[0061] S45, Targeting regional node o i Based on dynamic polar radius r′ i Velocity modulus s i , motion trend characteristics θ i , rate of change of motion trend Δθ i and local density features n i Calculate the local threat score:

[0062] μ i =σ(w1·f r (r′ i )+w2·f s (s i )+w3·f θ (θ i )+w4·f Δθ (Δθ i )+w5·f d (n i ));

[0063] Where, μ i For the local threat score, σ(·) is the Sigmoid normalized activation function, which compresses the local threat score to (0,1), f r (r′ i ) represents the dynamic polar radius r′ i Normalized feature processing result, f s (s i ) represents the velocity modulus s i Normalized feature processing result, f θ (θ i ) represents the trend characteristic θ i Normalized feature processing result, f Δθ (Δθ i ) represents the rate of change of the trend of motion Δθ i Normalized feature processing result, f d (n i) represents the local density feature n i The normalized feature processing results, w1, w2, w3, w4, and w5 are the weighted coefficients for the fusion of local threat scores;

[0064] S45, in each azimuth end region A j Internally, based on local threat score μ i Initialize the azimuth threat level matrix M j The local threat score μ i Fill the azimuth threat level matrix M according to the regional node index order. j Complete the directional threat level matrix M at the corresponding location. j The generation, and the azimuth end region A j Associated storage.

[0065] The system quantifies and analyzes the speed, direction, and distance characteristics of target objects within each directional region to calculate the directional threat level and construct a level matrix. Simultaneously, it predicts potential risk areas in the future time series by combining target trajectory trends, ultimately establishing a dynamic risk distribution model. This model enables early assessment and dynamic adjustment of potential collision risks for autonomous vehicles in multiple directions, significantly improving the response capability to sudden obstacles or high-threat targets. Through continuous updates to the level matrix, the system maintains sensitivity to dynamic changes in the environment, enhancing the foresight and defensiveness of obstacle avoidance strategies.

[0066] Optionally, S5 specifically includes:

[0067] S51, Based on the azimuth threat level matrix M j Extract each azimuth end region A j Local threat score of inner region nodes at time step t Current location velocity vector and movement trend characteristics Calculate the trajectory position within the next K time steps to generate a predicted trajectory set P. i :

[0068]

[0069] S52, Based on the predicted trajectory set P i Predicted locations of nodes in each region and the local threat score of the regional node at the current time step t Will This can be directly applied to every future time step t+k, and a future local threat score can be set.

[0070] S53. At each future time step t+k, based on the predicted trajectory set Pi Predicted locations of nodes in each region Local density features n i and future local threat score Fusion generation of dynamic risk distribution map R (t) :

[0071]

[0072] Among them, R (t+k) Let R(x,y,t+k) be the dynamic risk distribution map for each future time step t+k. (t+k) The risk intensity at each location point, O ′ For a subset of effective target objects, For the predicted trajectory set P i Predicted locations of nodes in each region The Euclidean distance to the reference point (x,y), where α is the weighting coefficient of the Euclidean distance influence and γ is the weighting coefficient of the local node density influence.

[0073] By integrating the directional threat level matrix with movement trend information, the system constructs a dynamic risk distribution map with time prediction capabilities. A risk evolution mechanism is introduced before path decision-making, ensuring that path planning is based not only on static risks but also on the trend of risk changes over a future period, thus strengthening the time dimension of path safety. Especially in multi-target interaction or dense environments, this mechanism can proactively avoid potentially high-risk intersection areas, effectively reducing the probability of path conflicts and planning failure rates, achieving dynamic obstacle avoidance capabilities oriented towards time windows.

[0074] Optionally, the RRT path optimization method that introduces dynamic risk costs specifically includes:

[0075] During the extended sampling process, the extended sampling points (x) s ,y s Limited to low-risk areas The low-risk region is defined as the region that satisfies R(x,y,t)≤δ, where δ is a set risk threshold.

[0076] During the path expansion process, based on the dynamic risk distribution map R (t) Initialize the dynamic risk cost path expansion tree T, and find and expand the sampling points (x). s ,y s The nearest node v in Euclidean distance near Generate an extended path segment, which points to an extended sampling point (x). s ,y s ), and calculate v near With new node v newThe Euclidean distance d(v) between them near ,v new ), and select the dynamic evaluation benchmark (x) in the extended path segment in real time. m ,y m The corresponding risk intensity R(x) m ,y m ,t), R(x m ,y m ,t) and d(v near ,v new Combining these elements, a basic dynamic risk cost C is formed. risk (v near ,v new ):

[0077] C risk (v near ,v new )=d(v near ,v new )×(1+ρ0R(x m ,y m ,t));

[0078] Wherein, ρ0 is the dynamic risk level weighting coefficient, which is used to adjust the response sensitivity of the extended path segment to changes in environmental risk;

[0079] Based on the predicted trajectory set P i Calculate the path cost correction factor Φ(x) m ,y m The path cost correction factor Φ(x,t) m ,y m ,t) is determined by the trajectory deviation TrjBias(x m ,y m ) and future local threat score Composed of weighted combinations:

[0080]

[0081] Where ρ1 is the trajectory deviation weighting coefficient and ρ2 is the threat level weighting coefficient;

[0082] Basic dynamic risk cost C risk (v near ,v new ) and path cost correction factor Φ(x) m ,y m The evaluation value Q(v) of the composite formation of the extended path segment. near ,v new ):

[0083] Q(v near ,v new)=d(v near ,v new )×(1+ρ0R(x m ,y m ,t))×(1+Φ(x m ,y m ,t));

[0084] During path expansion, the evaluation value Q(v) of the expanded path segment is used. near ,v new The principle of minimization is used to dynamically select the expansion direction and expansion base point, update the dynamic risk cost path expansion tree T, and continuously optimize the path expansion process by refreshing the basic dynamic risk cost status and path cost correction factor status in real time.

[0085] By introducing dynamic risk cost and path cost correction factors, the system establishes a composite path segment evaluation mechanism. This elevates the traditional path extension cost from a single Euclidean distance to a multi-factor weighted index that integrates risk, trajectory deviation, and threat level, enhancing the environmental awareness capability of path selection. This mechanism can assign differentiated safety weights to different path segments, guiding the path to avoid high-risk areas from the initial stage of extension. This improves the initiative of the extension strategy and the path convergence speed, making it particularly suitable for application scenarios with high requirements for path adjustment response speed in dynamic, high-frequency environments.

[0086] Optionally, S6 specifically includes:

[0087] S61, Based on dynamic risk distribution map R (t) The RRT path optimization method with dynamic risk cost is introduced to perform path expansion search. The current position (x0, y0) of the unmanned vehicle is set as the starting point of the path expansion search, and the dynamic risk cost path expansion tree T is initialized.

[0088] S62. Define the low-risk domain:

[0089]

[0090] Where R(x,y,t) is the dynamic risk distribution diagram R at each future time step t+k. (t+k) The risk intensity at each location point is defined by δ, which is a set risk threshold; in the low-risk region... Extended sampling is performed within the range, and extended sampling points (x) are randomly selected. s ,y s );

[0091] S63. In the dynamic risk cost path expansion tree T, find and expand the sampling point (x). s ,y s The nearest node v in Euclidean distance near and with v nearAs the base point, an extended path segment is generated, which points to the extended sampling point (x). s ,y s Select an intermediate reference point (x) in the extended path segment. m ,y m As a dynamic evaluation benchmark, the future trajectory and threat level of the target object are used as path cost correction factors, and real-time evaluation is performed.

[0092] Q(v near ,v new )=d(v near ,v new )×(1+ρ0R(x m ,y m ,t))×(1+Φ(x m ,y m ,t));

[0093] Where Q(v) near ,v new ) represents the evaluation value of the extended path segment, v new For the new node, d(v) near ,v new R(x) represents the Euclidean distance between the two nodes, ρ0 represents the risk level weighting coefficient, and R(x) represents the risk level weighting coefficient. m ,y m (x,t) is the intermediate reference point (x) in the extended path segment. m ,y m In the dynamic risk distribution map R (t) The corresponding risk intensity, path cost correction factor Φ(x) m ,y m ,t) is defined as:

[0094]

[0095] Where ρ1 is the trajectory deviation weighting coefficient, TrjBias(x m ,y m ) is (x m y m ) and the predicted trajectory set P i The deviation between the predicted locations of nodes in each region, where ρ2 is the threat level weighting coefficient. To predict the trajectory set P at each future time step t+k i Future local threat scores for nodes in each region;

[0096] S64, Evaluation value Q(v) based on extended path segment near ,v new Select the new node v that best represents the current extended path segment. newUpdate the dynamic risk cost path expansion tree T.

[0097] During path expansion, a dynamic risk-driven expansion search strategy, combined with low-risk domain sampling constraints and the principle of minimizing comprehensive path segment evaluation values, effectively improves the robustness and safety of the expanded path in dynamic environments. The system can filter high-risk areas in real time during the expansion tree growth process and dynamically adjust the expansion direction based on risk costs, significantly reducing the probability of the path falling into local extrema. Furthermore, this strategy, linked with the output of the Motion-DETR model, enables rapid response to newly emerging targets and path reconstruction, enhancing the system's real-time obstacle avoidance capability and path adaptability.

[0098] Optionally, S7 specifically includes:

[0099] During the path expansion search process, the cumulative cost of each sampling node is updated in real time, and the expansion direction is dynamically selected based on the principle of minimum cost to generate the optimal safe driving path in the dynamic environment.

[0100] S71, The new node v with the best current extended path segment quality. new As a new extension base point, the evaluation value Q(v) based on the extended path segment is used. near ,v new ), calculate the starting point v of the path expansion search. start to v new The cumulative path value C total (v new ):

[0101]

[0102] Wherein, P(v start ,v new ) for v start to v new A set of path segments;

[0103] S72. During the path expansion process, the cumulative path cost C of each expansion base point in the dynamic risk cost path expansion tree T is updated in real time. total (v new ), and with the current extended base point v near For reference, from v near Select the expansion base point v with the minimum cumulative path cost from the set of nearest scalable nodes. best This will serve as the next direction for path expansion.

[0104] S73, Comparison v new The cumulative path value C total (v new The sum of the cumulative path costs of the candidate parent node and the candidate parent node, if a candidate parent node v exists.alt Make it possible to pass through v alt If the cumulative path value is smaller, then update v. new The parent node pointer optimizes the connection relationship of the dynamic risk cost path expansion tree T;

[0105] S74. When the path extension reaches the endpoint node v global Or, if the expansion termination condition is met, from the endpoint node v global Starting from the parent node, backtrack backwards to the path expansion search starting point v. start The nodes are connected sequentially to generate the optimal safe driving path based on the minimum cumulative path cost in a dynamic environment.

[0106] The system introduces a dynamic update mechanism for cumulative path cost values ​​during the expansion search process and generates the final path according to the principle of minimum cumulative cost value during path backtracking, achieving a complete closed loop from "single expansion optimization" to "global path optimization". By continuously recording the cumulative cost value of each expansion node and adjusting the parent node structure, the system can effectively avoid the legacy effects of early low-quality paths and improve the global rationality of the final path. This mechanism is particularly suitable for scenarios that require frequent path adjustments in dynamic and risky environments, ensuring that each path output is optimal in the current environment, thus enhancing the system's stability and practicality.

[0107] The beneficial effects of this invention are:

[0108] This invention addresses the key challenge of directional object detection and obstacle avoidance recognition for unmanned vehicles in complex dynamic environments by constructing an intelligent discrimination system based on dynamic perception map modeling and path planning optimization. By collecting multi-source heterogeneous sensor data and performing synchronization, alignment, and standardization preprocessing in both temporal and spatial dimensions, a highly consistent environmental perception dataset is generated, ensuring the robustness and real-time performance of the perception input. In the feature extraction stage, the system introduces the Motion-DETR dynamic perception model, combined with target detection and trajectory association strategies, to achieve accurate extraction and structured modeling of different types of dynamic target objects, overcoming the problems of missed and false detections of targets in dynamic scenes caused by traditional static image perception methods.

[0109] Furthermore, this invention constructs a dynamic environmental perception map centered on the autonomous vehicle and maps environmental information to multiple directional regions using multi-directional segmentation rules. Combining the speed, direction, and other characteristic information of target objects within each region, it calculates the threat level and establishes a dynamic risk distribution model. This achieves spatial distribution modeling and forward warning of potential risk areas, effectively improving the perception capability of target movement trends and dynamic risk changes during the path planning stage. During path planning optimization, the system introduces a dual mechanism of dynamic risk cost and path cost correction factor, embedding the target object's movement trend deviation and threat level into the path cost calculation, dynamically adjusting the path expansion strategy. This mechanism significantly improves the system's path adaptability and safety in the face of uncertain environmental changes, overcoming the problem that traditional RRT path expansion methods cannot achieve global optimization in complex, obstacle-dense areas.

[0110] Furthermore, during the path search process, this invention constructs a path expansion tree through a dynamic update mechanism that accumulates path cost values, and dynamically optimizes the expansion structure using a minimum cost criterion, ultimately backtracking to generate the optimal safe driving path. Compared to traditional static path evaluation strategies, this scheme can integrate local environmental risks and global cost information in real time, achieving dynamic correction of the expansion structure and outputting the overall optimal path, thus improving the global rationality and local safety of the navigation path. Finally, the system outputs the optimal path as the driving command, providing highly safe and robust intelligent obstacle avoidance and path decision support for autonomous vehicles.

[0111] In summary, this invention constructs an end-to-end intelligent obstacle avoidance system architecture by integrating multi-source perception, dynamic graph modeling, path cost optimization, and intelligent expansion strategies. It solves the problems of low obstacle avoidance accuracy, slow response, and unstable paths in existing unmanned vehicle systems in dynamic multi-target environments. It has strong environmental adaptability, path feasibility, and deployment practicality, and significantly improves the navigation and decision-making capabilities of intelligent vehicle systems in complex real-world scenarios. Attached Figure Description

[0112] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0113] Figure 1 This is a schematic diagram of the intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of unmanned vehicles proposed in this invention.

[0114] Figure 2 This is an overall flowchart of the intelligent obstacle identification and obstacle avoidance method for autonomous object detection based on the orientation of unmanned vehicles proposed in this invention. Detailed Implementation

[0115] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0116] refer to Figure 1 The autonomous obstacle detection and avoidance intelligent identification system based on the location of unmanned vehicles includes the following modules:

[0117] The perception data acquisition module is used to collect multi-source perception data of the environment around the unmanned vehicle, perform time synchronization, spatial alignment and standardization preprocessing, and generate an environmental perception dataset.

[0118] The feature extraction module is used to input the environmental perception dataset into the Motion-DETR dynamic perception model, extract information about each target object in the environment, and generate a target object feature set.

[0119] The perception map construction module is used to construct a dynamic environment perception map based on the feature set of the target object, with the current position of the unmanned vehicle as the center, and divide the dynamic environment perception map into multiple directional end regions according to the preset orientation division rules.

[0120] The threat assessment module is used to calculate the threat level based on the information of the target object in each directional area, generate a directional threat level matrix, and predict the future movement trajectory by combining the movement trend of the target object to establish a dynamic risk distribution map.

[0121] The path expansion and optimization module is used to take the dynamic risk distribution map as input, set the current position of the unmanned vehicle as the starting point of path planning, and adopt a path expansion and optimization method that introduces dynamic risk costs to dynamically select the expansion direction.

[0122] The path cost optimization module is used to update the cumulative path cost of each expansion node in real time during the path expansion process, and dynamically optimize the expansion structure based on the principle of minimizing the cumulative path cost.

[0123] The path output module is used to backtrack from the endpoint node to the starting node along the parent node link to generate the optimal safe driving path in a dynamic environment.

[0124] refer to Figure 2 The method for autonomous object detection and obstacle avoidance based on the location of unmanned vehicles includes the following steps:

[0125] S1. Collect multi-source perception data of the environment around the unmanned vehicle, and perform time synchronization, spatial alignment and standardization preprocessing on the multi-source perception data to generate an environmental perception dataset.

[0126] S2. Input the environmental perception dataset into the Motion-DETR dynamic perception model to extract information about each target object in the environment and form a target object feature set.

[0127] S3. Based on the target object feature set, construct a dynamic environment perception map with the current position of the unmanned vehicle as the center, and divide the dynamic environment perception map into multiple directional end regions according to the preset orientation division rules.

[0128] S4. Within each azimuth region, calculate the local threat score based on the target object information and generate the azimuth threat level matrix.

[0129] S5. Based on the azimuth threat level matrix and the target object's motion trend information, predict the target object's future trajectory and establish a dynamic risk distribution map;

[0130] S6. Based on the dynamic risk distribution map, the current position of the unmanned vehicle is set as the starting point of the path planning. The RRT path optimization method with dynamic risk cost is used to expand the path search. The future trajectory and threat level of the target object are used as path cost correction factors to update the dynamic risk cost path expansion tree.

[0131] S7. During the path expansion search process, the cumulative cost of each node is updated in real time, and the expansion direction is dynamically selected based on the principle of minimum cost to generate the optimal safe driving path in the dynamic environment.

[0132] In this embodiment, S1 specifically includes:

[0133] S11. Collect multi-source perception data of the environment around the unmanned vehicle, including visual image data and three-dimensional point cloud data.

[0134] S12. Perform time synchronization processing on multi-source sensing data and extract the acquisition timestamp t of the visual image data. c The timestamp of the 3D point cloud data acquisition r Using the nearest neighbor matching strategy, select t c and t r Pair visual image data and 3D point cloud data with the smallest absolute time difference;

[0135] S13. Assign a unified synchronization timestamp to the paired visual image data and 3D point cloud data. The unified synchronization timestamp is set to the acquisition timestamp t of the visual image data. c The timestamp of the 3D point cloud data acquisition r The average value;

[0136] S14. Perform spatial alignment processing on the time-synchronized multi-source sensing data, based on the extrinsic parameter matrix T. extThe intrinsic parameter matrix K transforms the 3D point cloud data in the radar coordinate system to the camera coordinate system, and projects the transformed 3D point cloud coordinates onto the image plane to obtain the corresponding 2D pixel positions. Points that exceed the boundary of the image plane after projection are removed, thus realizing the fusion of visual image data and 3D point cloud data in a unified spatial reference system.

[0137] S15. Standardize the multi-source perception data after time synchronization and spatial alignment. The visual image data is normalized to standardize the pixel values ​​to the range [0,1]. The three-dimensional point cloud data is standardized according to the set ranging range to standardize the ranging values ​​to the range [0,1].

[0138] S16. The standardized multi-source sensing data is fused and organized according to a unified spatial indexing rule to generate an environmental sensing dataset.

[0139] In this embodiment, S2 specifically includes:

[0140] S21. Input the environmental perception dataset into the Motion-DETR dynamic perception model;

[0141] S22. Using a convolutional neural network set within the visual feature extractor module, the visual image data in the environmental perception dataset is encoded to extract visual spatial feature information and generate a visual feature tensor F. img ;

[0142] S23. Using a multilayer perceptron network set within the point cloud feature extractor module, the 3D point cloud data in the environmental perception dataset is encoded point-by-point to extract the spatial geometric features of the 3D point cloud and generate the point cloud feature tensor F. pc ;

[0143] S24. For the visual feature tensor F respectively... img and point cloud feature tensor F pc Introducing a self-attention weight matrix for feature reweighting:

[0144]

[0145] Among them, F′ img For the weighted visual feature tensor, F′ pc For the weighted point cloud feature tensor, Let be the self-attention weighting matrix corresponding to the visual feature tensor. Let be the self-attention weighting matrix corresponding to the point cloud feature tensor;

[0146] S25. Weighted visual feature tensor F′ img With weighted point cloud feature tensor F′ pcPerform additive fusion processing to generate a fused feature tensor through the fusion module:

[0147] F fusion =α·F′ img +(1-α)·F′ pc ;

[0148] Among them, F fusion Let α be the fusion feature tensor, and α be the fusion weight coefficient.

[0149] S26, Fen the feature tensor F fusion The input is fed into the dynamic feature modeling module, where a multi-head self-attention mechanism is used to model and fuse the dynamic relationships between feature units, outputting a target object feature set O. This target object feature set includes category information, center position coordinates, velocity vector, and motion trend features.

[0150]

[0151] Among them, o i Let c be the feature vector of the i-th target object. i For the category information of the i-th target object, (x i ,y i () represents the center coordinates of the i-th target object. Let θ be the velocity vector of the i-th target object in the x and y directions. i Let represent the motion trend characteristics of the i-th target object.

[0152] In this embodiment, S3 specifically includes:

[0153] S31. Based on the target object feature set O, extract the center position coordinates (x, y, y) of the target object. i ,y i ), velocity vector and motion trend characteristics θ i Using the current position coordinates (x0, y0) of the unmanned vehicle as a reference, the predicted position is calculated based on the center position coordinates and velocity vector of the target object and the set short-term prediction step size. Then, the dynamic polar radius is determined based on the Euclidean distance between the predicted position and the current position of the unmanned vehicle, and the dynamic polar angle is determined based on the directional relationship between the position of the target object and the unmanned vehicle after the short-term prediction.

[0154] S32. Based on the dynamic polar radius, select effective target objects that meet the sensing radius condition and construct a subset of effective target objects. ′ The perception radius is the maximum perception distance threshold set with the current position of the unmanned vehicle as the center, which limits the processing object to the physical space area that the unmanned vehicle can perceive and react to in real time.

[0155] S33. Centered on the current position of the unmanned vehicle, based on the effective target object subset O ′ A dynamic environment perception map is constructed, wherein the nodes of the dynamic environment perception map are effective target objects, and the node attributes include the predicted position, velocity vector and motion trend characteristics of the effective target objects.

[0156] S34. Based on the dynamic polar angles corresponding to the nodes of the dynamic environment perception map, and according to the preset direction division rules, divide the dynamic environment perception map into N azimuth end regions, each azimuth end region being numbered A. j , where j∈{1,2,...,N}.

[0157] In this embodiment, S4 specifically includes:

[0158] S41. Based on the dynamic environment perception map, for each azimuth end region A j Extract region nodes and construct a subset O of region nodes. j Among them, the region node subset O j Each region node o i This includes predicted position, velocity vector, motion trend characteristics, and dynamic polar radius;

[0159] S42, For each region node o i Stored velocity vector calculation region node o i velocity modulus s i ;

[0160] S43, Based on each region node o i Current motion trend characteristics θ i Movement trend characteristics compared to the previous moment Calculate the rate of change of motion trend Δθ i ;

[0161] S44, Statistical region node o i The number of neighboring nodes within the perception radius, n i As a local density feature;

[0162] S45, Targeting regional node o i Based on dynamic polar radius r′ i Velocity modulus s i , motion trend characteristics θ i , rate of change of motion trend Δθ i and local density features n i Calculate the local threat score:

[0163] μ i =σ(W1·f r (r′ i )+w2·fs (s i )+w3·f θ (θ i )+w4·f Δθ (Δθ i )+w5·f d (n i ));

[0164] Where, μ i For the local threat score, σ(·) is the Sigmoid normalized activation function, which compresses the local threat score to (0,1), f r (r′ i ) represents the dynamic polar radius r′ i Normalized feature processing result, f s (s i ) represents the velocity modulus s i Normalized feature processing result, f θ (θ i ) represents the trend characteristic θ i Normalized feature processing result, f Δθ (Δθ i ) represents the rate of change of the trend of motion Δθ i Normalized feature processing result, f d (n i ) represents the local density feature n i The normalized feature processing results, w1, w2, w3, w4, and w5 are the weighted coefficients for the fusion of local threat scores;

[0165] S45, in each azimuth end region A j Internally, based on local threat score μ i Initialize the azimuth threat level matrix M j The local threat score μ i Fill the azimuth threat level matrix M according to the regional node index order. j Complete the directional threat level matrix M at the corresponding location. j The generation, and the azimuth end region A j Associated storage.

[0166] In this embodiment, S5 specifically includes:

[0167] S51, Based on the azimuth threat level matrix M j Extract each azimuth end region A j Local threat score of inner region nodes at time step t Current location velocity vector and movement trend characteristics Calculate the trajectory position within the next K time steps to generate a predicted trajectory set P. i :

[0168]

[0169] S52, Based on the predicted trajectory set P i Predicted locations of nodes in each region and the local threat score of the regional node at the current time step t Will This can be directly applied to every future time step t+k, and a future local threat score can be set.

[0170] S53. At each future time step t+k, based on the predicted trajectory set P i Predicted locations of nodes in each region Local density features n i and future local threat score Fusion generation of dynamic risk distribution map R (t) :

[0171]

[0172] Among them, R (t+k) Let R(x,y,t+k) be the dynamic risk distribution map for each future time step t+k. (t+k) The risk intensity at each location point, O ′ For a subset of effective target objects, For the predicted trajectory set P i Predicted locations of nodes in each region The Euclidean distance to the reference point (x,y), where α is the weighting coefficient of the Euclidean distance influence and γ is the weighting coefficient of the local node density influence.

[0173] In this embodiment, the RRT path optimization method that introduces dynamic risk costs specifically includes:

[0174] During the extended sampling process, the extended sampling points (x) s ,y s Limited to low-risk areas The low-risk region is defined as the region that satisfies R(x,y,t)≤δ, where δ is a set risk threshold.

[0175] During the path expansion process, based on the dynamic risk distribution map R (t) Initialize the dynamic risk cost path expansion tree T, and find and expand the sampling points (x). s ,y s The nearest node v in Euclidean distancenear Generate an extended path segment, which points to an extended sampling point (x). s ,y s ), and calculate v near With new node v new The Euclidean distance d(v) between them near ,v new ), and select the dynamic evaluation benchmark (x) in the extended path segment in real time. m ,y m The corresponding risk intensity R(x) m ,y m ,t), R(x m ,y m ,t) and d(v near ,v new Combining these elements, a basic dynamic risk cost C is formed. risk (v near ,v new ):

[0176] C risk (v near ,v new )=d(v near ,v new )×(1+ρ0R(x m ,y m ,t));

[0177] Wherein, ρ0 is the dynamic risk level weighting coefficient, which is used to adjust the response sensitivity of the extended path segment to changes in environmental risk;

[0178] Based on the predicted trajectory set P i Calculate the path cost correction factor Φ(x) m ,y m The path cost correction factor Φ(x,t) m ,y m ,t) is determined by the trajectory deviation TrjBias(x m ,y m ) and future local threat score Composed of weighted combinations:

[0179]

[0180] Where ρ1 is the trajectory deviation weighting coefficient and ρ2 is the threat level weighting coefficient;

[0181] Basic dynamic risk cost C risk (v near ,v new ) and path cost correction factor Φ(x) m ,y mThe evaluation value Q(v) of the composite formation of the extended path segment. near ,v new ):

[0182] Q(v near ,v new )=d(v near ,v new )×(1+ρ0R(x m ,y m ,t))×(1+Φ(x m ,y m ,t));

[0183] During path expansion, the evaluation value Q(v) of the expanded path segment is used. near ,v new The principle of minimization is used to dynamically select the expansion direction and expansion base point, update the dynamic risk cost path expansion tree T, and continuously optimize the path expansion process by refreshing the basic dynamic risk cost status and path cost correction factor status in real time.

[0184] In this embodiment, S6 specifically includes:

[0185] S61, Based on dynamic risk distribution map R (t) The RRT path optimization method with dynamic risk cost is introduced to perform path expansion search. The current position (x0, y0) of the unmanned vehicle is set as the starting point of the path expansion search, and the dynamic risk cost path expansion tree T is initialized.

[0186] S62. Define the low-risk domain:

[0187]

[0188] Where R(x,y,t) is the dynamic risk distribution diagram R at each future time step t+k. (t+k) The risk intensity at each location point is defined by δ, which is a set risk threshold; in the low-risk region... Extended sampling is performed within the range, and extended sampling points (x) are randomly selected. s ,y s );

[0189] S63. In the dynamic risk cost path expansion tree T, find and expand the sampling point (x). s ,y s The nearest node v in Euclidean distance near and with v near As the base point, an extended path segment is generated, which points to the extended sampling point (x). s ,y s Select an intermediate reference point (x) in the extended path segment. m ,y mAs a dynamic evaluation benchmark, the future trajectory and threat level of the target object are used as path cost correction factors, and real-time evaluation is performed.

[0190] Q(v near ,v new )=d(v near ,v new )×(1+ρ0R(x m ,y m ,t))×(1+Φ(x m ,y m ,t));

[0191] Where Q(v) near ,v new ) represents the evaluation value of the extended path segment, v new For the new node, d(v) near ,v new R(x) represents the Euclidean distance between the two nodes, ρ0 represents the risk level weighting coefficient, and R(x) represents the risk level weighting coefficient. m ,y m (x,t) is the intermediate reference point (x) in the extended path segment. m ,y m In the dynamic risk distribution map R (t) The corresponding risk intensity, path cost correction factor Φ(x) m ,y m ,t) is defined as:

[0192]

[0193] Where ρ1 is the trajectory deviation weighting coefficient, TrjBias(x m ,y m ) is (x m y m ) and the predicted trajectory set P i The deviation between the predicted locations of nodes in each region, where ρ2 is the threat level weighting coefficient. To predict the trajectory set P at each future time step t+k i Future local threat scores for nodes in each region;

[0194] S64, Evaluation value Q(v) based on extended path segment near ,v new Select the new node v that best represents the current extended path segment. new Update the dynamic risk cost path expansion tree T.

[0195] In this embodiment, S7 specifically includes:

[0196] During the path expansion search process, the cumulative cost of each sampling node is updated in real time, and the expansion direction is dynamically selected based on the principle of minimum cost to generate the optimal safe driving path in the dynamic environment.

[0197] S71, The new node v with the best current extended path segment quality. new As a new extension base point, the evaluation value Q(v) based on the extended path segment is used. near ,v new ), calculate the starting point v of the path expansion search. start to v new The cumulative path value C total (v new ):

[0198]

[0199] Wherein, P(v start ,v new ) for v start to v new A set of path segments;

[0200] S72. During the path expansion process, the cumulative path cost C of each expansion base point in the dynamic risk cost path expansion tree T is updated in real time. total (v new ), and with the current extended base point v near For reference, from v near Select the expansion base point v with the minimum cumulative path cost from the set of nearest scalable nodes. best This will serve as the next direction for path expansion.

[0201] S73, Comparison v new The cumulative path value C total (v new The sum of the cumulative path costs of the candidate parent node and the candidate parent node, if a candidate parent node v exists. alt Make it possible to pass through v alt If the cumulative path value is smaller, then update v. new The parent node pointer optimizes the connection relationship of the dynamic risk cost path expansion tree T;

[0202] S74. When the path extension reaches the endpoint node v global Or, if the expansion termination condition is met, from the endpoint node v global Starting from the parent node, backtrack backwards to the path expansion search starting point v. start The nodes are connected sequentially to generate the optimal safe driving path based on the minimum cumulative path cost in a dynamic environment.

[0203] Example 1:

[0204] To verify the feasibility of this invention in practice, it was applied to autonomous driving tests conducted in a smart manufacturing park. In this park, unmanned vehicles face typical highly dynamic mixed scenarios, including intersecting pedestrians, various types of transport vehicles, autonomous robots, and temporarily stacked goods. These factors exhibit a complex combination of high dynamism, spatial uncertainty, and target diversity. Traditional static map-based and rule-based path planning systems struggle to operate stably in such unstructured environments, often experiencing frequent path interruptions, obstacle avoidance failures, and repetitive planning delays. This results in poor overall system stability, untimely response, and even potential safety hazards.

[0205] The project team selected four typical areas in the park for comparative testing: the main road, the warehouse intersection, the factory perimeter, and the warehouse driveway, representing a progressive scenario from low risk to high interference. The system of this invention is integrated and deployed on a general-purpose medium-sized unmanned logistics vehicle platform, equipped with high-precision LiDAR, millimeter-wave radar, and RGB-D binocular cameras, and works in conjunction with a high-performance edge computing module for online inference and path planning decisions.

[0206] In the operational process, the perception data acquisition module first synchronizes the time and aligns the spatial coordinates of multiple source sensors to construct a unified-scale environmental perception dataset. The Motion-DETR dynamic perception model performs real-time target recognition and feature extraction on the current scene, outputting the center point coordinates, velocity vector, type label, and attitude estimation information of each target, forming a target object feature set. The system constructs a dynamic environmental perception map with the autonomous vehicle as the pole and divides it into azimuth regions using a polar coordinate partitioning method. Threat level assessment is performed independently in each region, constructing an azimuth threat level matrix.

[0207] Based on the matrix and the predicted future motion trend of the target object, the system further develops a dynamic risk distribution model. This model maps the trend of risk evolution over time to a two-dimensional space, supports dynamic risk linkage analysis across multiple regions within a short time window, and provides prior conditions for subsequent path planning. During the path expansion phase, the system employs an improved RRT algorithm, integrating dynamic risk cost and path cost correction factor into the path cost calculation. This avoids the expansion direction from entering high-risk areas, improving path safety and convergence speed.

[0208] Each path extension is completed within 50ms. The system updates the cumulative path cost of the extended nodes in real time, executes the optimal parent node selection strategy, and backtracks to form the optimal safe path for the current moment after the extension is completed. If the dynamic risk distribution changes abruptly (such as a high-threat target suddenly approaching), the system will automatically trigger path reconstruction to ensure that it always stays on a feasible and low-risk operating track. The entire planning and backtracking process maintains high concurrency response and does not affect the synchronous execution of perception and path decision-making.

[0209] The data collected and organized during the experiment are shown in the table below.

[0210] Table 1. Performance comparison results of the present invention and traditional methods

[0211]

[0212]

[0213] The table above compares the core performance of the traditional method and the system of this invention in four typical test scenarios. In the scenario of a main road in a park with a low proportion of dynamic targets, the traditional method can basically complete the obstacle avoidance task, but it suffers from poor path stability, slow response, and high planning time. The system of this invention, on the other hand, achieves a 100% obstacle avoidance success rate, reduces planning latency by 41.5%, and exhibits excellent operational stability without any path reconstruction.

[0214] At complex intersections in warehouse areas and around factory buildings, where dynamic interactions are more frequent, the obstacle avoidance success rate of traditional methods drops to 85.4% and 78.1%, respectively. Path reconfiguration is performed 4-5 times, and the time consumed generally exceeds 65ms. The system of this invention, however, proactively avoids potentially high-risk paths through a dynamic risk guidance mechanism, maintaining an obstacle avoidance success rate above 91%, controlling the average path planning time to within 50ms, and significantly reducing the number of path reconfigurations. Overall, its path consistency and response efficiency are superior to traditional methods.

[0215] In the most restricted warehouse driveway scenarios, this invention effectively predicts path blocking trends through a dynamic threat level matrix in a high-density target environment, allowing for early detours to low-risk areas, increasing the obstacle avoidance success rate to 97.3%, and ensuring path reconstruction occurs less than once, effectively guaranteeing the continuity of path decisions and safety redundancy.

[0216] In summary, through field testing and verification in typical industrial parks with various road types and mixed dynamic target environments, the intelligent obstacle avoidance system based on autonomous vehicle orientation-based object detection in this invention significantly solves the core problems of traditional path planning methods, such as response lag, poor path stability, lack of risk perception, and insufficient path optimization granularity under dynamic target interference. Test results show that the system maintains a high obstacle avoidance success rate, low path reconstruction frequency, and stable path planning delay under different target densities, dynamic ratios, and scene complexities, demonstrating good path coherence and dynamic environment adaptability. Especially in complex areas with high dynamics and high-frequency interference, the system exhibits strong robustness and high stability, fully demonstrating the efficient collaboration between its perception, modeling, and decision-making modules. Based on the Motion-DETR perception model, dynamic risk distribution modeling, and the extended mechanism of introducing a path cost correction factor, this system outperforms existing technologies in terms of path generation quality, environmental perception depth, and overall resource efficiency. In summary, this invention not only excels in obstacle avoidance safety, system real-time performance, and structural flexibility, but also possesses excellent engineering feasibility and potential for widespread adoption. It is applicable to various intelligent driving application scenarios such as smart logistics, unmanned warehouse vehicles, and industrial park inspections, and has significant engineering promotion value and industrial transformation prospects.

[0217] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent obstacle avoidance and location-based object detection system for unmanned vehicles, characterized in that: Includes the following modules: The perception data acquisition module is used to collect multi-source perception data of the environment around the unmanned vehicle, perform time synchronization, spatial alignment and standardization preprocessing, and generate an environmental perception dataset. The feature extraction module is used to input the environmental perception dataset into the Motion-DETR dynamic perception model, extract information about each target object in the environment, and generate a target object feature set. The perception map construction module is used to construct a dynamic environment perception map based on the feature set of the target object, with the current position of the unmanned vehicle as the center, and divide the dynamic environment perception map into multiple directional end regions according to the preset orientation division rules. The threat assessment module is used to calculate the threat level based on target object information within each directional region, generate a directional threat level matrix, and predict future movement trajectories by combining the target object's movement trends to establish a dynamic risk distribution map, including: Based on the dynamic environment perception map, for each directional end area Extract region nodes and construct a subset of region nodes. Among them, the subset of regional nodes Each regional node This includes predicted position, velocity vector, motion trend characteristics, and dynamic polar radius; For each region node Stored velocity vector calculation region nodes velocity modulus ; Based on each regional node Current movement trend characteristics Movement trend characteristics compared to the previous moment Calculate the rate of change of the trend of motion ; Statistical region nodes Number of neighboring nodes within the perception radius As a local density feature; For regional nodes Based on dynamic polar radius Velocity Module Movement trend characteristics Rate of change of motion trend and local density features Calculate the local threat score: ； in, Score for local threats. Using a Sigmoid-normalized activation function, the local threat score is compressed to... , For dynamic polarity Normalized feature processing results For velocity modulus Normalized feature processing results Characteristics of movement trends Normalized feature processing results Rate of change of trend of motion Normalized feature processing results Local density features Normalized feature processing results , , , , The weighting coefficients for the local threat scores are used for fusion. In each azimuth end region Internally, based on local threat scores Initialize the azimuth threat level matrix Local threat score Fill the azimuth threat level matrix according to the regional node index order. Complete the directional threat level matrix at the corresponding location. The generation, and the azimuth end region Associated storage; The path expansion and optimization module is used to take the dynamic risk distribution map as input, set the current position of the unmanned vehicle as the starting point of path planning, and adopt a path expansion and optimization method that introduces dynamic risk costs to dynamically select the expansion direction. The path cost optimization module is used to update the cumulative path cost of each expansion node in real time during the path expansion process, and dynamically optimize the expansion structure based on the principle of minimizing the cumulative path cost. The path output module is used to backtrack from the endpoint node to the starting node along the parent node link to generate the optimal safe driving path in a dynamic environment.

2. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle as described in claim 1, characterized in that, The modules are connected in the following way: S1. Collect multi-source perception data of the environment around the unmanned vehicle, and perform time synchronization, spatial alignment and standardization preprocessing on the multi-source perception data to generate an environmental perception dataset. S2. Input the environmental perception dataset into the Motion-DETR dynamic perception model to extract information about each target object in the environment and form a target object feature set. S3. Based on the target object feature set, construct a dynamic environment perception map with the current position of the unmanned vehicle as the center, and divide the dynamic environment perception map into multiple directional end regions according to the preset orientation division rules. S4. Within each azimuth region, calculate the local threat score based on the target object information and generate the azimuth threat level matrix. S4 specifically includes: S41. Based on the dynamic environment perception map, for each azimuth end area Extract region nodes and construct a subset of region nodes. Among them, the subset of regional nodes Each regional node This includes predicted position, velocity vector, motion trend characteristics, and dynamic polar radius; S42, For each region node Stored velocity vector calculation region nodes velocity modulus ; S43, Based on each regional node Current movement trend characteristics Movement trend characteristics compared to the previous moment Calculate the rate of change of the trend of motion ; S44, Statistical Area Node Number of neighboring nodes within the perception radius As a local density feature; S45, Targeting regional nodes Based on dynamic polar radius Velocity Module Movement trend characteristics Rate of change of motion trend and local density features Calculate the local threat score: ； in, Score for local threats. Using a Sigmoid-normalized activation function, the local threat score is compressed to... , For dynamic polarity Normalized feature processing results For velocity modulus Normalized feature processing results Characteristics of movement trends Normalized feature processing results Rate of change of trend of motion Normalized feature processing results Local density features Normalized feature processing results , , , , The weighting coefficients for the local threat scores are used for fusion. S45, in each azimuth end region Internally, based on local threat scores Initialize the azimuth threat level matrix Local threat score Fill the azimuth threat level matrix according to the regional node index order. Complete the directional threat level matrix at the corresponding location. The generation, and the azimuth end region Associated storage; S5. Based on the azimuth threat level matrix and the target object's motion trend information, predict the target object's future trajectory and establish a dynamic risk distribution map; S6. Based on the dynamic risk distribution map, the current position of the unmanned vehicle is set as the starting point of the path planning. The RRT path optimization method with dynamic risk cost is used to expand the path search. The future trajectory and threat level of the target object are used as path cost correction factors to update the dynamic risk cost path expansion tree. S7. During the path expansion search process, the cumulative cost of each node is updated in real time, and the expansion direction is dynamically selected based on the principle of minimum cost to generate the optimal safe driving path in the dynamic environment.

3. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle as described in claim 2, characterized in that, S1 specifically includes: S11. Collect multi-source perception data of the environment surrounding the unmanned vehicle, including visual image data and three-dimensional point cloud data. S12. Perform time synchronization processing on multi-source sensing data and extract the acquisition timestamp of visual image data. Acquisition timestamp of 3D point cloud data Using the nearest neighbor matching strategy, select and Pair visual image data and 3D point cloud data with the smallest absolute time difference; S13. Assign a unified synchronization timestamp to the paired visual image data and 3D point cloud data. The unified synchronization timestamp is set to the acquisition timestamp of the visual image data. Acquisition timestamp of 3D point cloud data The average value; S14. Perform spatial alignment processing on the time-synchronized multi-source sensing data, based on the extrinsic parameter matrix. and intrinsic parameter matrix The system transforms the 3D point cloud data in the radar coordinate system to the camera coordinate system, and projects the transformed 3D point cloud coordinates onto the image plane through the intrinsic parameter matrix to obtain the corresponding 2D pixel positions. Points that exceed the boundary of the image plane after projection are removed, thereby achieving the fusion of visual image data and 3D point cloud data in a unified spatial reference system. S15. Standardize the multi-source sensing data after time synchronization and spatial alignment. Specifically, the visual image data is normalized by standardizing the pixel values ​​to the specified range. The 3D point cloud data is standardized according to the set ranging range, and the ranging values ​​are standardized to the interval. ; S16. The standardized multi-source sensing data is fused and organized according to a unified spatial indexing rule to generate an environmental sensing dataset.

4. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle as described in claim 2, characterized in that, S2 specifically includes: S21. Input the environmental perception dataset into the Motion-DETR dynamic perception model; S22. Using a convolutional neural network set within the visual feature extractor module, the visual image data in the environmental perception dataset is encoded to extract visual spatial feature information and generate a visual feature tensor. ; S23. By using a multilayer perceptron network set in the point cloud feature extractor module, the 3D point cloud data in the environmental perception dataset is encoded point by point to extract the spatial geometric features of the 3D point cloud and generate a point cloud feature tensor. ; S24. For the visual feature tensor respectively and point cloud feature tensor Introducing a self-attention weight matrix for feature reweighting: ； ； in, For the weighted visual feature tensor, For the weighted point cloud feature tensor, Let be the self-attention weighting matrix corresponding to the visual feature tensor. Let be the self-attention weighting matrix corresponding to the point cloud feature tensor; S25, Weighted visual feature tensor With weighted point cloud feature tensor Perform additive fusion processing to generate a fused feature tensor through the fusion module: ； in, To fuse feature tensors, For fusion weighting coefficients; S26, Fusing Feature Tensors The input is fed into the dynamic feature modeling module, where a multi-head self-attention mechanism is used to model and fuse the dynamic relationships between feature units, outputting a target object feature set. The target object feature set includes category information, center position coordinates, velocity vector, and motion trend features: ； in, For the first The feature vector of a target object For the first Category information of each target object, For the first The center coordinates of the target object For the first The target object is , velocity vector in the direction, For the first The motion trend characteristics of a target object.

5. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle as described in claim 2, characterized in that, S3 specifically includes: S31, Based on target object feature set Extract the center position coordinates of the target object Velocity vector and movement trend characteristics Using the current coordinates of the autonomous vehicle For reference, based on the center position coordinates and velocity vector of the target object, the predicted position is calculated according to the set short-time prediction step size. Then, the dynamic polar radius is determined based on the Euclidean distance between the predicted position and the current position of the unmanned vehicle, and the dynamic polar angle is determined based on the directional relationship between the position of the target object and the unmanned vehicle after the short-time prediction. S32. Based on the dynamic polar radius, select effective target objects that meet the sensing radius condition and construct a subset of effective target objects. The perception radius is the maximum perception distance threshold set with the current position of the unmanned vehicle as the center, which limits the processing object to the physical space area that the unmanned vehicle can perceive and react to in real time. S33. Centered on the current position of the unmanned vehicle, based on a subset of effective target objects. A dynamic environment perception map is constructed, wherein the nodes of the dynamic environment perception map are effective target objects, and the node attributes include the predicted position, velocity vector and motion trend characteristics of the effective target objects. S34. Based on the dynamic polar angles corresponding to the nodes of the dynamic environment perception map, and according to the preset direction division rules, divide the dynamic environment perception map into... Each directional end region is numbered as follows: ,in, .

6. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle as described in claim 2, characterized in that, S5 specifically includes: S51, Based on the azimuth-based threat level matrix Extract each azimuth end region Inner region nodes at time step Local threat score Current location Velocity vector and movement trend characteristics To carry out the future Trajectory position estimation within each time step, generating a predicted trajectory set. : ； S52, Based on Predicted Trajectory Sets Predicted locations of nodes in each region and regional nodes at the current time step Local threat score ,Will Directly extend to every future time step And set future local threat scores. ; S53, in every time step in the future Based on the predicted trajectory set Predicted locations of nodes in each region Local density features and future local threat score A dynamic risk distribution map is generated by fusion. : ； in, For every step of the future Dynamic risk distribution map For every step of the future Dynamic risk distribution map The risk intensity at each location point in the middle, For a subset of effective target objects, For the set of predicted trajectories Predicted locations of nodes in each region With reference point Euclidean distance, The weighting coefficients are the influence coefficients of Euclidean distance. The weighting coefficients are influenced by the local node density.

7. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle as described in claim 2, characterized in that, S6 specifically includes: S61, Based on dynamic risk distribution map The RRT path optimization method with dynamic risk cost is introduced to expand the path search and determine the current position of the autonomous vehicle. Set as the starting point for path expansion search, and initialize the dynamic risk-cost path expansion tree. ; S62. Define the low-risk domain: ； in, For every step of the future Dynamic risk distribution map The risk intensity at each location point in the middle, The set risk threshold; in the low-risk zone Extended sampling is performed within the range, and extended sampling points are randomly selected. ; S63, In the dynamic risk cost path expansion tree In the process of finding and expanding sampling points Euclidean nearest node and with As an extension base point, an extension path segment is generated, which points to the extension sampling point. Select intermediate reference points in the extended path segment. As a dynamic evaluation benchmark, the future trajectory and threat level of the target object are used as path cost correction factors, and real-time evaluation is performed. ； in, The evaluation value for the extended path segment, For the new node, Let be the Euclidean distance between the two nodes. This is a risk level weighting coefficient. intermediate reference points in the extended path segment In the dynamic risk distribution map The corresponding risk intensity and path cost correction factor. Defined as: ； in, The weighting coefficient for trajectory deviation. for With the set of predicted trajectories The deviation between the predicted positions of nodes in different regions. As a weighted coefficient for threat level, For every step in the future Predicted trajectory set Future local threat scores for nodes in each region; S64. Evaluation value based on extended path segments Select the new node that best suits the current extended path segment. Update the dynamic risk cost path expansion tree .

8. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle according to claim 7, characterized in that, The RRT path optimization method that introduces dynamic risk costs specifically includes: During the extended sampling process, the number of sampling points is extended. Limited to low-risk areas Internally generated, the low-risk domain is defined as satisfying The area, in which The set risk threshold; During the path expansion process, based on the dynamic risk distribution map Initialize the dynamic risk cost path expansion tree Find and expand sampling points Euclidean nearest node Generate extended path segments, which point to extended sampling points. and calculate With new nodes Euclidean distance between Real-time selection of dynamic evaluation benchmarks in extended path segments Corresponding risk intensity ,Will and Combined, they form a basic dynamic risk cost. : ； in, It is a weighting coefficient for dynamic risk levels, used to adjust the sensitivity of extended path segments to changes in environmental risks; Based on predicted trajectory set Calculate the path cost correction factor The path cost correction factor From trajectory deviation Score of future local threats Composed of weighted combinations: ； in, The weighting coefficient for trajectory deviation. This is a weighting factor for the threat level; Basic dynamic risk cost and path cost correction factor Evaluation value of compound formation of extended path segment : ； During path extension, the evaluation value of the extended path segment is used. The principle of minimization is used to dynamically select the expansion direction and expansion base point, and to update the dynamic risk-cost path expansion tree. Furthermore, the path expansion process is continuously optimized by updating the status of basic dynamic risk costs and path cost correction factors in real time.

9. The intelligent obstacle identification system for autonomous object detection and avoidance based on the orientation of an unmanned vehicle according to claim 2, characterized in that, Specifically, S7 includes: During the path expansion search process, the cumulative cost of each sampling node is updated in real time, and the expansion direction is dynamically selected based on the principle of minimum cost to generate the optimal safe driving path in the dynamic environment. S71, The new node with the best current extended path segment. As a new extension base point, the evaluation value based on the extension path segment. Calculate the starting point of the path expansion search. to Cumulative path value : ； in, for to A set of path segments; S72. During the path expansion process, the dynamic risk-cost path expansion tree is updated in real time. Cumulative path value of each extended base point and based on the current expansion base point For reference, from Select the expansion base point with the minimum cumulative path cost from the set of nearest scalable nodes. This will serve as the next direction for path expansion. S73, Comparison Cumulative path value The sum of the cumulative path costs to the candidate parent node; if a candidate parent node exists... Make it possible to pass If the cumulative path value is smaller, then update Parent node pointers, optimizing the dynamic risk cost path expansion tree The connection relationship; S74. When the path extension reaches the endpoint node Or, if the extended termination condition is met, start from the endpoint node. Starting from the parent node, backtrack backwards to the starting point of the path expansion search. The nodes are connected sequentially to generate the optimal safe driving path based on the minimum cumulative path cost in a dynamic environment.

Images