结合小波域去噪和运动估计的监控场景分析研究-通信与信息系统专业论文.docxVIP

结合 结合小波域的去噪和运动估计的监控场景分析研究 SURVEILLANCE SCENE ANALYSIS USING DENOISING AND MOTION ESTIMATION IN WAVELET DOMAIN ABSTRACT With the development of video monitoring technology, processing of surveillance video has become a research hotspot. Surveillance video often contains some noise due to changes in weather; low-light and so on. Noise not only worsen video quality and affect people’s subjective feeling, but also enlarge the transmission bandwidth and data storage space, that bringing a lot of inconveniences for further processing. In addition, monitoring scene is temporally stationary, video sequences has strong temporal correlation, the data generated by camera are very large, so increases data transfer and storage costs. To resolve these problems, this paper mainly researches wavelet de-noising and motion estimation in monitoring scene. Traditional video de-noising methods can debase image noise, but destroy image edge information. Wavelet transform has good time-frequency characteristics and has become an efficient mathematics tool applied to image de-noising. According to the noise type in surveillance video, the paper takes advantages of the characteristics of wavelet threshold de-noising method and wiener filtering, propose an image de-noising algorithm. Compared with wavelet transform threshold de-noising algorithms or spatial wiener filtering, the proposed method can improve PSNR efficiently. Video sequence in monitoring scene is different from the head-shoulder image sequence; moving objects may occupy only a small part of the whole image. If we do motion estimation for the entire image, it will not only time-consuming but also laborious. This paper proposes an algorithm, that applying moving object detection into wavelet domain, and combining with motion estimation, that reducing unnecessary search and saving storage space. The experimental results prove that this approach decrease the computational complexity obviously compare with wavelet motion estimation