关键词: 畜牧机器人, AprilTag, 建图与定位, 单目

Abstract: In order to obtain the location information of animal husbandry autonomous mobile robots in largescale breeding farms, a map building and location algorithm based on TWODIMENSIONAL code (AprilTag) was proposed. A monocular camera on the robot was used to observe the AprilTag in the environment. The state of the robot was estimated based on extended Kalman Filter (EKF) by integrating the measurement of the robot motion with the wheel odometer. On this basis, the pose of each AprilTag in the global coordinate system was calculated. In order to restrain the increase of cumulative errors in the process of building the graph, the estimated results were taken as the initial values before the loop, and the leastsquares problem with the projection relation was constructed, and the LevenbergMarquardt (LM) method was used to solve the leastsquares problem. The global optimization was used to eliminate the cumulative error when the loop was detected. Experimental results showed that the proposed algorithm can build the AprilTag map in a large range, and the average error and maximum error of robot localization using the map were 0.075 86 m and 0.303 93 m, respectively.

Key words: animal husbandry robot, AprilTag, mapping and positioning, monocular