Research On Depth Information Acquistion Of Robot Binocular Vision Based On Phase Primitives Of Wavelet

Robot binocular vision, as a most active branch in robot technology research, is an important feature of intelligent robots. Depth information acquisition is not only one of its important research purposes, but also a key step to reconstruct 3-D environment. Consequently, further research on robot binocular vision provides necessary technological support for solving problems such as the navigation of intelligent robots, the orientation of industrial robots.A complete algorithm of robot binocular vision contains six main parts: image acquisition, camera calibration, feature extraction, image matching, depth calculation and depth information interpolation, among which camera calibration and picture matching are widely studied. This dissertation mainly makes further exploration on the key step of depth information acquisition—image matching. Binocular vision image matching is to find imaging locations of the tie point in stereo image pairs. The dissertation introduces parallel binocular vision theory in details and provides the mathematic model of image matching. And further more, it discusses four important factors of image matching: the selectivity of matching primitives, the confirmation of matching principles, matching constraint and the structure of matching algorithm. In this dissertation, we choose dual-tree complex wavelet as the matching primitive with the reason that it has good shift independence, directional selectivity, complete reconstruction, high computing efficiency and excellent local characteristics owned by wavelet. Many researchers have done a lot of work on complex wavelet in image matching, but traditionaly they only make use of one-direction primitive. In this paper six-direction phase primitives are used to abtain more texture information. Combined with pyramid structure of dual-tree complex wavelet, the dissertation adopts relaxation algorithm.