Path Planning Of Unmanned Vehicle Based On Binocular Vision Perception

Unmanned vehicles with a high level of autonomy and automation help human to conduct cumbersome,monotonous or dangerous tasks.It is widely used in traffic safety,social life,aerospace,national defense and military fields.The improvement of the automation level of unmanned vehicle depends on its own sensors,especially the environment perception sensor.Autonomous navigation of unmanned vehicle based on machine vision is a research focus of unmanned vehicles.Binocular vision is capable of sensing the distance of objects,which is the main way for humans and most organisms to realize environmental perception and navigation.Binocular vision navigation is an important research direction for unmanned vehicles,and its difficulty lies in the real-time performance of binocular vision perception processing algorithm and path planning algorithm with excellent performance.This paper mainly aims at lower cost and higher processing speed of unmanned vehicle hardware platform construction,the binocular vision obstacle perception with hardware acceleration algorithm,and the path planning of the vehicle.To deal with the large computation demand in binocular vision processing and path planning algorithm,this paper proposes an unmanned vehicle system based on FPGA+ARM hardware architecture.The flexibility of ARM together with the parallel acceleration ability of the FPGA make the system have high performance for unmanned vehicles application.The binocular vision system is built for the rapid visual acquisition,real-time stereo correction of the current scene image information.FPGA is used to realize parallel acceleration operation of stereo matching algorithm and real-time obstacle recognition.On the ARM,the path planning algorithm under the constraint information of obstacles is performed.Finally,experiments verify the effectiveness and feasibility of the unmanned vehicle hardware system.In the process of obstacle detection,the operation including stereo matching,binarization,median filtering and connected domain detection are carried out for the left and right gray images after their synchronization and correction.To balance the performance and computation cost,semi global stereo matching algorithm is adopted to realize disparity estimation and obstacle distance calculation.The main contribution is to use the pipeline principle and parallel operation of FPGA to accelerate the calculation speed.In the programming process,the flow design between parallel calculation within a module,merging repeated comparison blocks and pipeline operation are used to improve the calculation efficiency and reduce the resource occupancy rate.Based on the limited hardware resources of unmanned vehicle,the real-time obstacle detection within 2 meters is realized,which meets the real-time and accurate motion control requirements of the unmanned vehicle in this paper.In the process of path planning,aiming at the problem of target unreachable and local minimum problem in the traditional artificial potential field method,the novel method based on virtual target points is proposed solve the local minimum problem in traditional methods.Combined with the improved repulsion potential function,the problem of unreachable targets can be effectively eliminated.The effectiveness and superiority of the proposed algorithm are proved by simulation results.Finally,the improved artificial potential field algorithm is applied to the constructed unmanned vehicles to verify the feasibility and improvement of the proposed algorithm.The main contributions of this paper include two aspects:（1）On the basis of limited resources,a FPGA implementation framework of semi-global stereo matching algorithm based on block comparison and pipeline processing is proposed to realize the fast real-time perception of the depth information of the current scene.（2）Aiming at the local minimum problem of the traditional artificial potential field method,a novel artificial potential field method based on virtual target points is proposed to avoid the local minimum.