一种有效的实时人群计数方法

对固定镜头下视频序列中运动人体的检测和跟踪方法进行研究，利用灰度图像差分双向投影信息检测人体目标，提出一种基于统计运动区域几何特征固定比例的分割算法，使用最近邻匹配方法对人体进行跟踪。完整地实现了一个有效的实时人群计数系统。大量室内和室外场景实验结果表明，该算法具有很好的实时性(每秒处理25帧~30帧且可并行处理4路视频)、对光照变化的鲁棒性以及对稀疏人群检测精度高等特点。

Effectual Real-time Method for Crowd Counting

This paper discusses the detection and tracking of moving human in video sequences gathered from a fixed camera, using a target detection algorithm based on projection information of gray value. An effectual algorithm is proposed to segment multiple people. In tracking process, this paper proposes an algorithm called nearest neighbor matching and implements a whole system for crowd counting. Large numbers of experimental results on several real video sequences from outdoor and indoor scenes have demonstrated the real-time effect(25~30 frames/s and implementing parallelization with four videos), a high precision for sparse crowd and robustness for various environments change.