A calibration system for vision-guided agricultural robots☆

Abstract Vision-guided robots have many potential applications in agriculture. Using a vision system to guide the robot allows a lower accuracy requirement for the robot, leading, hopefully, to cheaper robotics. However, it is important that the relationship between robot co-ordinates and vision co-ordinates is known. The vision system must be able to convert the position of an object in the image into a position that the robot can understand. This paper describes a system to calibrate the robot-camera combination. The robot draws a grid which is analysed by the vision system. This includes all sources of distortion in the calibration loop. The algorithm can allow for variable lighting conditions, and uses knowledge of the expected grid to search for each grid point. A piecewise-linear map is fitted to the points and the resultant mapping is stored. The system is fast, robust and easy for the unskilled operator to use.