Research Of3D Video Depth Map Processing And ASIC Design

3D (Three dimension,3D) video is the important direction in the ifeld of digitalmedia，which in a stage of vigorous development. The competition of the developmentand application becoming more and more fierce in recent years. Compared to thetraditional2D video,3D video can provide the depth information of the observedobjects, allowing the user to feel three-dimensional visual expeirence.3D video canpromote digital media and creative industires to upgrade and birnging new opportunitiesfor traditional digital video technology in the field of aerospace military, education,health care, entertainment. The development of Free Viewpoint Video technique is oneof the most promising fields in the free viewpoint3D video process. Compared toconventional planar video processing technology, three-dimensional video has hugeimage data processing, storage, trafifc, and their algorithms and system design is verycomplex. With the multiview stereo video sequence processing, it is inevitable thatbirng the huge data processing bandwidth. For storage media and transmission media,they are unable to accept. The basic method of free viewpoint video processingtechnology is to estimated and reconstruction of multiview video based on the depthmap, which can significantly reduce the data bandwidth, so the process and optimize ofdepth map is particularly important which can reduce the amount of data in order toachieve real-time free viewpoint3D video processing.Depth map is calculated based on disparity offset values. Disparity vector can bepredicted by the combined temporal and interview estimation method which is morepractical for hardware circuit design and realization, because the motion estimationresources can be reused as much as possible, with decreasing processing databandwidth, computation time and resources. So the combined temporal and interview iswidely used in multiview video processing technology. While combined temporal andinterview estimation method is based on block search within a specific window andwith specific cost tfinction, depth map directly calculated from the block based disparityprediction result has much block effect and prediction noise, especially at largebackground or continuous areas. To compress depth map more eiffciently,optimization method need to be developed to reduce such prediction noises.The adaptive color image segmentation method can handle the depth image noise prediction effectively. On the basis of adaptive color segmentation method, this paperusing ASIC to complete the estimation and enhancement of depth map. The system isvalidated by simulation, the results show that: the depth image processing method hasthe advantages of real time, compatibility, practical features, suitable for real-time freeviewpoint video3D processing.