Special Vehicle Path Planning Considering Terrain Characteristics Research On Methods For Identitying Driving States

In recent years,with the rapid growth of China’s economy and the rapid advancement of urbanization,the construction of national major projects has been increasing.In order to meet the transportation tasks of various materials and ensure the safety of personnel and property during transportation,this article discusses and studies the optimal path planning problem of special vehicles and the identification problem of vehicle driving status.In the study of path planning for special vehicles,this paper proposes an optimal path planning implementation plan for special vehicles under complex road conditions,taking into account various influencing factors such as road slope and road curvature.In terms of vehicle driving state recognition,this article is based on deep learning object detection,object tracking and other image recognition detection technologies.The vehicle driving state recognition is divided into two sub tasks:vehicle target detection and object tracking,and then the recognition of vehicle driving state is studied in detail as follows:（1）In the research of special vehicle path planning,this article uses a forward star data structure to store real road network data.By extracting parameters such as size,load,wheelbase,and maximum deflection angle of special vehicles,a special vehicle model is constructed,and the minimum turning radius Rmin and minimum channel width W of the vehicle are obtained.Then,the trafficability function Pi was used to determine the accessibility of special vehicles at bridge tunnel and turning nodes.Finally,the optimal path planning for special vehicles was determined using different road resistance weights.The comprehensive optimal path not only considers the length of the road segment（η_l）Travel time（η_t）The number of bridges and tunnels(S_qs)and the degree of intersection（J_l）,as well as the combined influence of multiple factors such as terrain features such as average road slope（S_a）and road curvature（J_a）on the planning path results,were also considered.The research results were integrated into a special vehicle path planning system based on the JS language platform for visual display and analysis.（2）In terms of vehicle object detection research,this article considers the limitations of traditional object detection algorithms and conducts research and experiments on the current popular deep learning based object detection algorithms.Finally,based on the YOLOv5algorithm of One Stage,in order to address the issues of missed detection and false detection due to occlusion in the YOLOv5 algorithm,we have improved the original GIo U Loss to CIo U Loss,Adding a pyramid pooling module（PPM）to the backbone network and an attention mechanism（CBAM）to the detection network were improved.Finally,a self-made special vehicle dataset was used for training and validation.The improved model improved the overall accuracy of m AP by 2.51 percentage points,which can to some extent solve the problem of"special car"type vehicles missing detection,proving the practicality of the improved algorithm.（3）In the research of vehicle driving status,combined with the improved YOLOv5algorithm and Deep SORT algorithm,the improved YOLOv5 is used as the detector for Deep SORT target tracking,and road monitoring videos are taken as experimental data.Through the detection and tracking of video vehicles,the driving plan and predicted trajectory of the vehicles are used to identify the driving status of vehicles such as straight,reverse,lane changing,and turning,Finally,based on Py Qt5,a system was designed that can track and recognize the driving status of vehicles in daily surveillance videos.