Research On Automatic Navigation Of Forest Road Based On Machine Vision

In recent years,China has increased its investment in road construction in forest areas,and the crisscrossed cement forest roads provide convenience for autonomous navigation.However,the road conditions of forest roads are complex,and there will be weeds,mud,rolling stones,etc.At the same time,with the changes of seasons and weather,the color and texture of vegetation and soil around forest roads vary greatly.These conditions give forestry equipment to forest roads.Visual navigation on the Internet brings great difficulties.Therefore,in this paper,the experimental platform of the rear-drive front steering structure is the research object,and the research of forest visual navigation is carried out.The main research contents are:(1)The simple linear iterative clustering algorithm was not sensitive to image texture,so a method of adding texture distance in the clustering process was proposed,obtaining image texture information through local binary patterns,and integrating it into the calculation formula of Euclidean distance.At the same time,in order to reduce the impact of different illumination on visual navigation and improve the accuracy of super pixels contours fitting the road edge,this paper studied the color components of forest road images under three different illumination intensities,and improved the calculation method of color distance,it was more in line with the environmental characteristics of forest roads.The test results showed that the super pixels was obtained by improving simple linear iterative clustering algorithm,its shape was more regular,and the edge fitted better.(2)The classification result of the support vector machine affects the navigation effect,1283 road super pixels and 463 background super pixels were extracted,the color and texture feature vector of each super pixel,as the sample data of the support vector machine was calculated.In this paper,the grid search method and the cross-validation method were used to optimize the penalty coefficient C and the kernel parameter g of the support vector machine,and the optimal parameters C=6.9644 and g=4.5948 were obtained.The C and g were input into the support vector machine model for classification verification,and the classification accuracy rate was 94.387%.Compared with the classification accuracy rate of the unoptimized support vector machine,the accuracy rate was increased by 4.009%.(3)The test platform needs to obtain pose parameters in real time,firstly,forest road images were divided into regions for every 10 rows of units,and calculated the average value of the horizontal and vertical coordinates of the road pixels in each area by the centroid method,This average value was used as the center coordinate point of the road in the area.Secondly,because of the large changes in road curvature in forest areas,this paper proposed to use cubic B-spline curve to fit the center point of the road as the navigation path.The experimental results showed that this method had high accuracy in extracting the navigation path.According to the conversion relationship between the reference coordinate systems in the imaging process of the camera,the pose geometric model of the mobile platform during navigation was established.Through experimental verification,the geometric model obtained the pose parameters of the mobile platform to meet the navigation requirements.(4)The research results of this paper can meet the operation requirements of automatic navigation in the forest environment,conform to the direction of intelligent development of forestry equipment,and have certain reference value for visual navigation of forestry equipment.