Calibration of a pole-mounted camera using a neural network

The intelligent soft arm control (ISAC) system is a robotic aid for the physically challenged. ISAC's current task is to feed the physically challenged. To accomplish this task ISAC needs to recognize the objects on the table, pick up the utensil it needs and bring it to the user's mouth. A camera is being used to recognize the objects on the table and a pair of cameras is being used to locate the user's mouth in real time. This report describes the calibration of the vision system dedicated to object recognition. The camera used in this system is mounted on a pole which is attached to the base of the robot. This setup makes the ISAC system compact as well as allowing the camera to view more than just one particular location, since it rotates with the base of the robot. Once an object is located in the image, the position of the object must be converted to robot coordinates so that the robot can use it. Unfortunately, the image plane is not parallel to the table in this setup. This causes the mapping from the image coordinates to the robot coordinates to be nonlinear. This nonlinearity was the motivation behind using a neural network to "learn" the mapping from image coordinates to robot coordinates