运动人体图像分割算法研究

许多监视系统要求能对运动人体的行为进行监视,如监狱、仓库、保密室、办公室、会场等场景,对运动人体的识别特点是区分运动人体与其他运动目标,而非判别人的个体间差别.以帧间差分法分割运动目标图像为基础,通过人的外形模式匹配、肤色匹配和姿态匹配的融合构造了一种识别运动目标中人体运动目标的算法(AMHBR),该算法的特点是简洁高效、适合嵌入式系统等资源受限的监视系统.通过试验证明,该算法能在不同的场景中分割出运动人体图像,具有很好的鲁棒性和准确度,达到了实用的程度.该算法对于监视系统中自动控制镜头对准运动人体具有实际意义.

Study on image segmentation algorithm of moving human body

The video surveillance systems are demanding to monitor moving human behaviors, specifically for such environments as jails, warehouses, secret rooms, offices and auditoriums. Accordingly, it is an imperative to differentiate moving human bodies from other moving objects rather than make a difference amongst human bodies. In this study, the frame difference method is employed for image segmentations of moving targets. By matching such human features as appearances, colors and postures, an algorithm for moving human body recognition (AMHBR) is established to detect moving targets efficiently. In addition, this algorithm can be adapted for resource-restricted cases e.g. embedded systems. Based on experimental results, it is found that images of moving human bodies have been robustly and accurately segmented from different backgrounds. Therefore, this algorithm can be pragmatically applied for video surveillance systems via automatically controlling lenses towards moving human bodies.