Study On The Disparity Estimation Of The Stereo Vision

With the development of computer technology, video codec technology has greatly improved to some extent. However, single-view video information can not meet the demand of people's desire in the video information. In recent years, stereo video technology with 3D visual function and multidimensional viewpoint video technology has been paid great attention to, and become a research hotspot.According to the statistics, about 60%-75% of the information that human obtained from outside is from visual system. Therefore, in order to meet the increasing demand on the visual system, it is bound to develop a series of advanced technologies such as information processing and handling which is in close relationship with the video information. For the video information processing and transmission, the primary problem is how to compress the video effectively. As we all know, the stereo video contains three-dimensional depth information of the scene. Therefore, the characterization of the natural features is much more real. However, in the stereo video, the amount of data is much larger than single-channel video. Therefore, an efficient stereo video compression is particularly important. Meanwhile, with the development of computer technology, stereo vision theory will be widely used in automatic navigation, industrial measurement, virtual reality, biomedical, military reconnaissance and other fields. In the stereo video, disparity estimation is a key point of the study, which has a very important role in the whole system. In this dissertation, we do some research on the disparity estimation in the stereo video.As the traditional dynamic programming algorithm implements the global energy minimum only in the scan line, the obtained disparity map has the obvious"fringe"phenomenon. We improve the global energy function and strengthen the restriction between the scanning lines, so the"fringe"phenomenon has been improved. We also study the disparity estimation algorithm in the belief propagation. By reducing the energy function and processing of the final disparity map, the accuracy has been improved.