Research Of Binocular Vision-based On Agricultural Robot Navigation System

Agricultural robot needs to identify crops when navigating autonomously,how to detect crop information real-time,fast and accurate,the utilization of information as navigation control of agricultural robot has become a hot topic in agricultural production.Meanwhile,some objective factors are usually complex and changeable,it's difficult to effectively establish visual navigation information of crop feature under the field environmental conditions.The navigation system in this thesis based on binocular vision technology,which is expected to provide in real-time and effective to navigation parameters for agricultural robot.The main content of the thesis including:(1)An overall scheme of robot navigation system is setup.(2)A parallel binocular vision optimization calibration model is proposed,and experimental results indicate that when the calibration distance is 190cm,the calibration error reaches the lowest with the abscissa 0.115pixel,and the ordinate 0.105pixel.And then Bougeuet algorithm is adopted to deal with the image rectification issue.Genetic algorithm is used to optimize neural network in order to calibrate cameras by using rectified image pairs.(3)A filtering method of binary crop image is proposed,which combines open operation and self-adaptive threshold statistical filter,in order to obtain valuable parts of green crops.Experimental results show that this filter method reduced the noise in binary image up to 17.63%.(4)An improved feature matching method of crop image is proposed,which combines with RANSAC algorithm greatly retained feature points of crop and implement features matching.Experimental results show that SURF-RANSAC method is effective under different crops and backgrounds,with matching rate of 96.9%,and the processing time is 0.92s with the standard deviation 0.03s.For an agricultural robot in low speed,this method meets the requirements of constructing three-dimensional information in real-time.(5)A navigation control strategy is designed based on three-dimensional information of crops.Simulating and experimental results show that maximum lateral deviation is 7.9cm with the average 2cm,and the maximum heading angle is 4.3° with the average 1.2° in the indoor.