摘要：The measurement value of the traditional binocular parallax distance is the distance between the reference point P and the center of the baseline of the binocular camera, and for the agriculture robot, because of the needs of ground operations, the cameras are usually installed in certain height from the ground with a certain angle with the horizontal direction, when we have to know the horizontal distance from the navigation reference point to the robot body and thus the next travel pose of the robot can be controlled by real-time. Obviously, the traditional binocular parallax distance measuring methods will no longer apply to this. In this regard, a new method for solving agricultural robot navigation reference point distance measurement is proposed. First, conduct calibration for the binocular system with the improved BP neural network, and secondly, obtain the left and right image coordinates of the navigation reference point (U1,V1) (U2,V2) with the improved SIFT features and input the BP neural networks trained in the calibration, and finally, output the coordinates of the navigation reference point in the world coordinate system (X ， Y), and then the horizontal distance between the navigation reference point and the robot body can be expressed as 22 .. S X Y . Experiments show that by this method, the maximum deviation of the actual field experiment test is 0.479cm, with the minimum deviation of 0.032cm, accuracy up to 99%, consuming a total of 55ms. And compared to the traditional binocular parallax distance ranging procedure, the computation is significantly reduced, with certain engineering practicability and feasibility.
关键词：Reference point range; Field navigation; Agricultural robot; Traditional ; binocular parallax; BP neural networks
