1. School of Electrical and Information Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China； 2. Nantong Guangyi Electromechanical Co., Ltd., Nantong, Jiangsu 226000, China
Abstract： To reduce pose estimation error of agricultural machinery by the GPS signal losing in the complex orchard environment, a multisource fusion localization method was proposed based on factor graph. RealSense D435i was integrated with color and depth camera and IMU to sense the surrounding environment of unmanned sprayer. Through the improved extraction method of adaptive threshold uniform Harris corner, the pyramid LK optical flow algorithm was used to match and track corner, and the camera motion was estimated by ICP algorithm. Based on the optimization algorithm，the pose estimation errors were iteratively optimized. The experiment was conducted on a forest road. Compared with inertial integrated navigation, the improved method could effectively estimate the sprayer pose in real time when GPS signal was lost. The experiment distance was 112.39 m in total, and the position update frequency after multisensor fusion was about 50 Hz. The results show that the RMS error of sprayer position is reduced by 0.816 m, and the maximum error is reduced by 4.613 m with the maximum error of attitude reduced by 2.713°.
