Study On Navigation Line Extraction Algorithm Of Agricultural Vehicle Based On Machine Vision

Automatic navigation technology of agricultural machinery is an important part of precision agriculture,and the extraction of navigation line is the primary task of automatic navigation.In recent years,image processing technology has been gradually applied in extraction of navigation line for its low cost.Therefore,it is necessary to study a fast,accurate and robust navigation line extraction algorithm.Based on the basic theory of image processing technology and the detection technology of crop row centerline,three key technologies of agricultural machinery navigation line extraction are studied: farm image preprocessing,crop row anchor point extracting and crop line fitting.Finally,the effectiveness of the algorithm is verified in the laboratory environment.The main work of this thesis can be summarized as follows:(1)A denoising algorithm for farmland image based on super-pixel block density is studied.Based on the research of traditional gray-scale algorithm,binarization algorithm and morphological filtering denoising algorithm,and aiming at the problem that the size of morphological operator in morphological denoising is not easy to determine and the effect of denoising is not thorough.The combination of super-pixel segmentation algorithm,traditional gray-scale and binarization results is considered in this thesis.By combining the super-pixel segmentation algorithm with the Super-Green feature gray-scale algorithm,the farmland image denoising algorithm based on the super-pixel block density is proposed in this thesis.Finally,a simple result analysis is made between the commonly used morphological denoising algorithm and the algorithm in this thesis.The denoising results show that the performance of this algorithm is better than that of morphological method,and can also be used as an effective denoising method for farmland images.(2)An algorithm for extracting crop line anchor points based on contour search is studied.Through the simulation of commonly used crop anchor point extraction algorithms,the advantages and disadvantages of each algorithm are analyzed.In order to solve the problems of inaccurate detection and poor real-time performance of commonly used crop anchor point extraction algorithms,using bar graph to detect crop contours,and screening the detection results according to certain thresholds are considered in this thesis.In the experiment,the commonly used crop anchor point extraction algorithms are evaluated qualitatively and quantitatively.The analysis results show that this algorithm can effectively overcome the speed and accuracy of the traditional algorithm,and can extract the anchor points of crop rows effectively.(3)An algorithm of crop row centerline detection based on contour search and line scanning is studied.Because crop row line is easy to be affected by the detection accuracy of the anchor point when using the anchor point for line detection.And under normal circumstances,the density of crops is much higher than that of weeds.Thus,in the foundation of the anchor point extraction algorithm of crop rows based on contour search,a improvement algorithm of a crop row line detection based on the density of green crops is considered in this thesis.The endpoint below a straight line is obtained by searching the contour in a certain strip area at the bottom of the image,and the endpoint above a straight line is searched at the top of the image.The optimal line where the crop row is located is determined according to the feature points near the line formed by the two points.Finally,the qualitative and quantitative evaluation of the traditional line detection algorithm and that of this algorithm are carried out in this thesis.The final results show that the algorithm is fast and accurate,taking less time,and the accuracy of line detection is high in the case of normal weeds and crop scarcity.(4)In the laboratory environment,a simple algorithm verification platform based on embedded system is built to validate the effectiveness of crop row centerline extraction algorithm based on contour search and line scanning.In order to verify the effectiveness of crop row line detection algorithm based on contour searching and line scanning,two green crop rows are simulated in the laboratory environment,and a car mounted camera is built to simulate agricultural vehicle.In the experiment,image processing results are transmitted to the slave computer for vehicle steering control.According to the established mathematical model between the vehicle and the camera,the lateral deviation of the head position is calculated,the processing time of each frame image and the transfer parameters to the slave computer are recorded.Through the analysis of vehicle running status and experimental results,the algorithm of crop row anchor point extraction based on contour searching and line scanning have faster processing speed and higher real-time in embedded system.