Abstract: As the algorithm of motion detection in video surveillance system is easily being affected by movements and noises in video background, and the result of salient motions is not accurate enough, an improved algorithm based on H.264 encoding is proposed to solve it. The algorithm is based on the Sum of Absolute Difference(SAD) of encoded macro-blocks, which are produced in the motion estimation process of H.264 encoding, then calculates its variance as the indicator for motion, and an auto-adjust threshold is used in the algorithm at the same time to decrease the effect of movements and noises in background. Experimental results show that the detection can decrease the effect of background movements and noises and make the salient motion detection more accurate, and keep the property of low complexity, which is 18.3% of frame differ algorithm, to make itself more suitable for real-time video surveillance system.

Key words: video surveillance, H.264 encoding field, motion detection, salient motion, Sum of Absolute Difference(SAD), real-time

摘要： 视频监控中运动目标检测算法易受背景、噪声等因素的影响，导致运动检测准确性降低。针对该问题，提出一种基于H.264编码域的显著运动目标检测改进算法。该算法基于编码过程中帧间运动预测模块输出像素值的绝对差值和，计算其标准差以作为运动判据，同时自适应设置运动判断阈值，以降低背景和噪声的影响。实验结果表明，相比于传统的低复杂度算法，该算法能有效降低背景和噪声等因素的影响，提高显著运动目标的检测准确性。相比于传统的运动检测算法，该算法保持了计算复杂度低的特点，算法复杂度仅为帧间差分法的18.3%，更适用于实时性要求高的监控系统。

关键词: 视频监控, H.264编码域, 运动检测, 显著运动, 绝对差值和, 实时性

CLC Number: 

• TP301.6