Abstract
It is required that a picture image is processed in realtime for the position measurement of a robot vehicle. This paper describes the method which detects artificial patterns of slit code with high precision and speed. Slit codes are constructed by arranging slits so that the intervals among their center lines are of definite lengths. Detection of a slit code is done in two steps. In the first step, each slit pattern is detected by a finite automaton which inputs sampled data of two-valued image in the direction of the length. In the second step, the projective invariant of the visual angles among center lines of four neighbouring slits is computed from the slit pattern. If the invariant value is zero, then the pattern is accepted as a subsequence of a slit code. Otherwise they are rejected.
The detection method is implemented in a system which scans the robot's environment by a TV camera through a rotating periscope. The experiments show its high reliability, that is, it accepts slit codes and rejects other noise slit patterns. Also the system makes it possible to follow a specified slit of the slit code using the method of slit pattern detection, while the robot is moving. Thus visual feedback is attained.
