Instant range finder: Based on trinocular vision method.

摘要

This thesis implements a Trinocular Vision method to improve two dimensional interpretation of a 3-D scene by means of correspondence and depth recovery. One aspect of Artificial Intelligence, in robots, involves shape, color, size, and depth interpretation. Binocular stereo was developed to mimic this human sense. However, this digital reference has problems with occlusion where the correspondences of a left and right image do not always match; hence the estimation of depth recovery is not 'perfect'. To improve these problems of Binocular Stereo a third camera was added; hence Trinocular Vision. With no access to resources of calibrated Trinocular images, an experiment was setup with three cameras to capture useful images. Due to orientation and position of these cameras a geometric constraint was used to develop this theory of Trinocular Vision. To implement these ideas, a Java script algorithm was developed to run an assimilation. The results produced by this assimilation can be seen in numerical interpretation, visual and graphical data. This data helped to demonstrate the improvement of correspondences and more reliable depth recovery of a 3-D environment. The accuracy of the experimental results varied between 0 -- 17% when compared to the measured data.