Practical stereo vision and multi-laser scanning in object face detection and orientation determination

Abstract

A laser-ranger-stereo-vision system to identify the pose of objects is described in this paper. The system consists of point-laser-sources and two deflection mirrors driven by two stepper motors. Two CCD cameras are used to capture stereo images of the object. Parallel processing is used to extract laser spots from both images, and to calculate the pose of the object. The system is semi-automated, and is controlled from a PC/SUN combination with two T800 transputers for image processing and six transputers for robot control. In this work, every effort has been made to make processing time a minimum, by imposing constraints to limit the search to a roughly evaluated area, and by using multiple switchable laser sources. A method to improve the range accuracy through fusion of multiple lasers and their a priori fixed relations is also given. The cost and accuracy of this system make it useful for a wide range of applications.