摘要
为改善视频监控中人流量检测的准确性问题,提出一种利用头发颜色特征的人流量跟踪检测方法。该方法对输入图像同时做如下两个操作:基于头发颜色特征的二值化和基于混合高斯模型的前景提取。对这两者合并后的结果做特征判别就可以得到人头区域。以人头区域为目标进行跟踪,分析运动轨迹特征后可以判断行人的进出方向及数目。实验表明该方法在双向行走、行人密集及背景干扰条件下均有很高的正确率。并且,每帧图像平均处理时间只需20 ms,完全可以满足实时处理的要求。
In order to improve the accuracy of the people counting detection in video surveillance,a people tracking and detection method based on hair color feature is proposed in this paper. The input images are processed with the following two operations at the same time：binarization based on hair color feature and foreground extraction based on Gaussian mixture model. Through analyzing the merging results of the two operations,head regions can be got. By tracking head regions and trajectory analysis,the moving direction and number of visitors will easily come to the judgment. The experimental result shows that the method has the high-accuracy even in the conditions of bidirectional moving,dense crowd or background interference. Further-more,the average processing time for per frame is only 20 ms. It can fully meet the requirements of real-time processing.
出处
《现代电子技术》
2014年第7期92-97,共6页
Modern Electronics Technique
基金
中国科学院战略性先导科技专项基金资助项目(XDA06020202)
上海市科委科研计划基金资助项目(12511501700):基于视频传感器融合的城市公共安全物联网关键技术研究
关键词
人流量实时检测
人头识别
目标跟踪
颜色空间
混合高斯模型
people counting real-time detection
head recognition
object tracking
color space
Gaussian mixture model
