Recognition And Coding Of ROI Based On Intelligence Surveillance

Because of the rapid development of multimedia technology and Internet, peoplecould enjoy rich and colorful information via different kinds of video-related services,such as Video Surveillance, Video Conference, 3D TV, Network TV, etc. Accordingly,in order to decrease the bandwidth consumption and storage space caused by these ser-vices, many research institutes and companies have conducted advanced research on videocoding for a long time and have drafted several video compression standards already.Recently, the developments of computer vision technology and computer graphics tech-nology have brought some new viewpoints to this realm. These new ideas drive us toconcentrate on the more important information within video data and thereby improvingthe coding efficiency. The ROI (Region of interest) related technology is a typical and hotone among these new ideas.ROI related technology is proposed based on characteristic of human visual systemthat people pay different attention to different parts of a video frames. Human eyes aremore sensitive to the noise,high brightness and some distortions of special areas. There-fore, an excellent video data compression method should guarantee the visual quality ofROI in the first place. This thesis researches bring the saliency theory of static image tovideo sequences in order to detect the ROI for every single frame. Then, in the phase ofROI coding, we could treat the ROI we detected before and balance the distortion of ROIand Non-ROI, to make sure ROI get finer quality. There are three phases in this researchwork listed in the following below:1. ROI Detection: We use the spectral residual and phase spectrum method to gen-erate saliency map for each frame of input video sequence. In the same time, we makea improvement on the spectral residual method to use phase spectrum to save about onethird computational complexity.2. ROI Pre-process: We first calculate the spatio-temporal saliency map, then usesome image processing technology to generate ROI in each frame and mark the ROIgrammar field in order to prepare for ROI coding phase.3. ROI Coding: We treat ROI specially and use gradient quantize to encode ROI.Then, ROI quality adjustable rate control algorithm is proposed to solve the problem of false alarm rate. This algorithm could adjust the ROI quality with the condition ofacceptable quality of non-ROI.The experiments results present that the average Hit Rate is between 0.45 and 0.65compared with Hand-labeled method, moreover, the SSIM of Coding results improve alot to prove that our method could automatically detect the ROI of video sequence andencode them with different QP. This method is a huge improvement compared to previousmethod, which uses manual marking ROI for the encoding demands. The method canfurthermore enhance the ROI quality in the limitation of bit rates and bandwidth.