A fast extraction primitive algorithm for implementation in a 3-D vision sensor

Using the stereovision principle, an intelligent 3-D vision sensor is currently under development for applications in autonomous vehicle or robot control. An algorithm integrates two major constraints: real time processing and autoadaptability. Based on a practical heuristic, the proposed method gives highly satisfactory results considering the computational times obtained (i.e. 16 ms to segment one image line) and the obvious autoadaptivity aspect after testing with a number of different image types.<>