Research On Field Road Perception And Path Tracking Based On Binocular Vision

The area of cultivated land in hilly regions,accounting for 63.2% of the total area in China,is an important crop production base for China's grains,oils and tobacco.However,poor condition of roads in the area of hilly regions results in the problems of large investment in labor force,poor security and low level of development.Lacking of labor force,hilly regions urgently need a transport vehicle with high degree of automation and safety.In this paper,the characteristics of no lane line,blurred boundary and complicated background of field roads in hilly area were analyzed.Binocular vision identifying the road ahead of the transport vehicle,three-dimensional matching and 3D reconstruction technology obtaining the three-dimensional information of the field road were used to calculate the local path tracking on the road in the field,and realize the automatic driving on the field road.The main research contents are as follows:(1)Binocular theory and calibration of binocular camera.The relationship between five basic coordinate systems and each coordinate system of binocular vision system were analyzed.According to the imaging model of monocular and binocular camera,the calibration method of Zhang Zheng You based on MATLAB toolbox was selected.Taking the black and white checkerboard as reference,the corners of checkerboard were extracted as features.The parameters of internal and external of the binocular camera were obtained.The results of different calibration distances were compared and analyzed.(2)Field road image processing algorithm.On the basis of analyzing the results of four different color spaces of RGB,Lab,HSI and HSV on the field road images,the HSV color space more adaptable to the field road conditions was selected for image processing.The V component in the HSV color space was separated for Otsu threshold segmentation and post-processing to get the road area and non-road.To solve the shadow problem existing in the road,the appropriate parameters were selected by experiments according to the feature that S component in the HSV color space is insensitive to shadows.Then S and V components' point operation and weighted average were carried out to highlight the shadow feature.Finally,shadows were extracted by Otsu threshold segmentation,and were added to the non-shadowed road area to get a complete road information.(3)Field road stereo information perception and path tracking.According to the feature of no obvious characteristics of the field road and the small difference in the gray value of the road area,based on the image processing,the statistical feature of the road's centroid point was extracted as the base element of the stereo matching.The least square method was used to smooth the statistical characteristics and then retaked the points to solve the problem of error or lack of path information caused by the unfavorable factors such as shadow and water stain.Later,the polar line constraint and homography matrix were applied for accurate matching and 3D reconstruction.Based on these,path tracking was present in the XY plane based on 3D information of filed road,and the problem of the convergence of the rolling window was solved by the method of partial rotation angle fusion of the front wheel.(4)Field road experiment.In order to verify the accuracy and feasibility of road image processing,three-dimensional information perception and path tracking algorithm in the field,the hardware system of road video acquisition,image processing and vehicle master controller were set up,the data transmission among three was built on the field road transport vehicle developed in the early stage.According to the 3D path information processed by the image processor,the main controller controls the steering-engine to realize the automatic driving of the transport vehicle.The results show that:(1)In HSV color space,the threshold segmentation of V component has a good recognition effect for ordinary field roads,roads with weeds and water stains,and non-hardened roads,but it is sensitive to roads with shadows,which causing partial road information loss.(2)For the shadow of different color depth and area,the shadow detection algorithm based on the S component and V component can complete the shadow detection well with simple algorithm and high operation efficiency.(3)The stereo matching algorithm based on homography matrix under condition of polar constraint can match and reconstruct the different roads statistical characteristics better.The matching elements are few and representative,and the matching speed is fast.The matching errors and 3D reconstruction errors of different road statistical features are no more than 10%.(4)In the actual test,Field road transport vehicle can better realize road perception and path tracking in the field,and always keep driving around the middle line of the road.Under linear path,the average deviation between the path track and the actual midline is 0.031 m,and the maximum deviation is 0.133 m.Under multi-curvature complex path,the average deviation is 0.069 m,and the maximum deviation is 0.195 m.Under undulating path,the average deviation is 0.105 m,and the maximum deviation is 0.216 m.It can meet the error requirement of the automatic driving transport vehicle on the field road.