The Study Of Video Coding Method On Surveillance Videos

Along with the emphasis on emergency situations, public safety, video surveillance has become a booming industry. Video compression technology as its core technology has become one of the hotspots. However, the existed video coding technology doesn’t consider the characteristics of video surveillance enough. There is still room for surveillance video coding method to improve.This paper firstly introduces the characteristics of surveillance video and demands for surveillance video coding, and then analyzes the disadvantage of existing coding techniques for surveillance videos. Based on the analysis above, we propose some new coding methods for surveillance videos, including the following aspects:Compared with other kinds of multimedia videos, surveillance videos show a prominent feature of background redundancy which results from the relatively static background. To full use this, this paper proposes an adaptive background-frame based coding method for surveillance videos. In this algorithm, the adaptive background image is used as a long-term reference frame to provide better prediction for exposed background blocks. The background image is generated and updated dynamically by using layer information and reconstructed images. During the updating period, the background area is also inter-frame filtered which contributes to better prediction for noisy videos. Over the AVS2test model RD9.2, this method can averagely save10.42%of the coding bit-rate, equivalent to0.25dB performance gain on surveillance videos.In order to achieve the motion information mentioned above, this paper studies the motion detection technique and region description method. The information of motion area is obtained by codebook model based motion detection method. The results are described by improved Pattern Quad-tree and finally written into the bit stream. Experiment results show that the motion detection and description method adopted in this paper presents the motion area well and doesn’t produce too much additional bitrate or consuming much computing resource. Another characteristic of surveillance videos is that noise always exists in these videos, especially in low light situation. This paper proposes a filtering method of reference frame called the motion-detection based adaptively synthesized reference frame, which is aiming at providing better motion estimation.for noisy videos. To preserve the picture details in motion area, we use two adaptive coefficients to update foreground and background area separately. Over the AVS2test model RD9.2, this method can save averagely12.42%of the coding bit-rate, equivalent to0.27dB performance gain on the noisy surveillance videos.