基于肤色分割的视频人体检测

摘要

For pedestrian movement shapes changeful and directions uncertain in video image sequence, an algorithm is presented to detect human body in video image sequence based on skin-tone segmentation. First, the moving target range will be located through background difference oftbe video image sequence. The background is self-adapted and obtained through an improved frame difference algorithm. And then the skin-tone area will be segmented in YCbCr color space to obtain the face region which is an important part of the body within the moving targets scope through elliptical skin model based on self-adaption luminance and some necessary post-processing such as morphological operations and some restrictive conditions. The experimental results show that the proposed method can detect the human body in video image sequence effectively. Compared with the similar kind of methods, the proposed method has the lower time complexity and can meet the requirements of real-time video processing better.%针对视频图像中人体运动形态多变、方向不确定等问题,提出了一种基于肤色分割的视频人体检测算法.采用改进的帧间差分方法求得自适应背景,与背景差相结合进行运动目标粗分割,在粗分割范围内进行肤色分割.在肤色分割时,利用YCbCr色彩空间的自适应亮度分段椭圆肤色模型及形态学操作、约束性条件等后处理确定人脸区域,得到人体目标.实验结果表明,该方法能有效地检测视频中的人体目标,与同类方法相比,该方法的时间复杂度更低,因而能更好地满足视频处理的实时性要求.