基于数学形态学和HSI颜色空间的人头检测

为提高视频监控中人数统计的实时性,提出一种人头检测算法。该方法在HSI(hue, saturation intensity)颜色空间建立人头颜色模型,并结合头发的灰度范围进行二值化。采用数学形态学的膨胀腐蚀算法进行预处理,并在边缘检测之后再次进行膨胀腐蚀,从而得到若干个候选区域。利用轮廓的几何特征综合判断这些区域是否为人头。实际监控的实验结果表明:该方法能够快速有效地检测出人头,很好地满足了视频监控中人数统计的实时性要求。

Head detection algorithm based on mathematical morphology and HSI color space

An approach of head detection was proposed to meet the requirement of the reducing people counting time in real-time video monitoring. A hair color model was established in HSI color space, and the image binarization was processed by considering the range of hair gray-level. The mathematical morphology dilation and erosion preprocessing algorithm was presented. Then, dilation and erosion algorithm was used again after image edge detection in order to get several candidate areas. At last, the geometrical characteristics of contour were used to judge whether these areas head or not. Experimental results showed that this algorithm was fast and effective, which could meet the real-time needs of intelligent video surveillance.