Vision Object Detection And Path Planning Algorithm For Unmanned Vehicle

With the rapid development of the economy,the scale of urban transportation network is becoming bigger than ever before,the traffic flow is increasing rapidly.Traffic congestion and traffic safety becomes severe problems to be solved in modern cities.Driven by science innovation,the technology of unmanned vehicle is developing rapidly in recent years.Unmanned vehicle is considered as a potential way to address these problems in the future.Nowadays this field is also a hot spot for the development of related industry.Object detection based on machine vision and path planning are two important components in the framework of unmanned vehicle systems.It is of both practieal value and theoretical significance to study and realize the real-time detection of traffic signs and autonomous path planning in complex environment for unmanned vehicles.In the field of traffic signs detection for unmanned vehicles,various regression-based deep learning network models are studied.After the discussion on principles and main techniques of several YOLO algorithms,the technical advantages of YOLOv3 algorithm are analyzed deeply.Three network models are built under the Darknet platform,and a number of experimental tests are carried out using a constructed data set of traffic sign pictures.It is verified that the YOLO algorithms can realize the location and classification of traffic signs in the picture simultaneously.By comparison and analysis using some evaluation indexes,it shows YOLOv3 algorithm has higher accuracy and better real-time performance than other algorithms when seven kinds of common traffic signs are tested.For unmanned vehicles path planning problem,the shortcomings of basic rapidly-exploring random tree(RRT)algorithm are analyzed.The idea of target-oriented search is introduced into the basic RRT algorithm.A target-oriented RRT algorithm for unmanned vehicle path planning is proposed,by which the distance of planned path is shortened and search efficiency is improved.The revised RRT algorithm fully utilizes the information of the target point during global search.After local path optimization,global path optimization and curve fitting,a safe and feasible path that satisfies the vehicle kinematics constraint is derived.The effectiveness and feasibility of the algorithm are verified by simulation and real vehicle test.For autonomous movement of an unmanned smart car in a scene built in the laboratory,the traffic sign detection based on Yolov3 algorithm and target-oriented RRT algorithm for path planning are applied to the unmanned smart car for experiments.Based on C++and Matlab/Simulink,the algorithms are coded to realize the functions of traffic sign detection and path planning for unmanned smart cars.After joint debugging of the programs,some experiments are performed in different scenarios and the validity and feasibility of the studied methods are verified.