Research On Inter-row Environment Perception And Navigation Method Of Tea Garden Weeding Robo

Tea is an important economic and green crop.Weed control in tea gardens is a crucial management process,and one that is labor-intensive and time-consuming.In the context of the current reduction in rural labor,the development of intelligent tea garden weeding robot is of significant importance for enhancing the level of tea garden management and improving the economic benefits of tea gardens.Environmental perception and navigation technology are core technologies of intelligent agricultural machinery.In addition to navigation along the operational path,the tea garden weeding robot requires perception and detection of obstacles and operational boundaries within the operational environment.In this paper,research is conducted on the environmental perception and navigation methods of tea garden weeding robots,with a specific focus on the aforementioned needs.The main research contents are as follows:(1)Based on the analysis of tea garden planting patterns and weed control machinery characteristics,the overall design of the weeding robot was completed.The environmental perception module was selected based on its requirements for environmental perception and navigation.The ZED stereo camera was chosen as the perception sensor,and the imaging principle and depth perception principle of the stereo camera were analyzed.Camera calibration was also completed.(2)A visual-based inter-row path navigation method for tea gardens was proposed according to the environmental characteristics of the tea garden operation path.This method preprocessed the camera-acquired images through multi-scale Retinex algorithm,bilateral filtering,H-component grayscale,and other steps.Based on the maximum inter-class variance method,the grayscale image was threshold segmented,and a two-step denoising method was designed to remove holes and noise in the binary image.After extracting the edge information of the inter-row road based on Canny edge detection,the road midpoint was obtained using a row scanning method and the road centerline was fitted as the navigation path.Experimental results showed that the accuracy of the method in segmenting the inter-row road area reached 87%,and the average heading angle deviation and lateral deviation of the fitted navigation line were 4.79° and 18.78 pixels,respectively.(3)To detect obstacles ahead of the weeding robot that may affect its operation,a method combining preliminary detection and SVM model judgement was designed.This method achieved preliminary obstacle detection based on the differences in color,hue,and saturation features between obstacle and non-obstacle regions in the image.Then,based on the HOG and LBP feature information of the obstacles and the SVM model,the detection and judgement of obstacles was realized.Experimental results showed that the obstacle detection rate of the algorithm reached 89.07%,which met the requirements for weeding robot operation.(4)The research on the stereo matching method of binocular images was conducted to obtain depth information of images.Firstly,the matching principles of local matching BM algorithm and semi-global matching SGBM algorithm were analyzed.In order to improve the matching accuracy of the SGBM algorithm,the cost calculation method was improved,and a TPAD-Census cost calculation method was proposed.The BM,SGBM,and improved SGBM algorithms were tested using the Middlebury dataset of binocular images.The results showed that the improved SGBM algorithm reduced the mismatch rate by 18.54% on the basis of slightly improving the matching time,and demonstrated better matching performance.(5)The fast ranging algorithm was designed to quickly obtain depth information of obstacle areas based on the improved SGBM algorithm.This algorithm improves matching efficiency through optimization of cost aggregation and disparity calculation.Experimental results showed that the average error of obstacle distance measurement using this algorithm is 7.82%,with an average measurement time of 969.01 ms,which met the obstacle positioning requirements of the weeding robot.(6)Based on the improved SGBM algorithm for obtaining depth information from images,a threshold segmentation method was used to detect the boundary between tea rows and the ground based on the depth difference.The boundary position was detected using a row scanning method,and the distance between the weeding robot and the tea row boundary was calculated based on the depth value at the midpoint of the boundary detection line.Experimental results showed that the detection accuracy of this algorithm for tea row boundaries was greater than 90%,and the accuracy of boundary position measurement was greater than 85%,which met the requirements of tea row boundary detection for the weeding robot.