基于AdaBoost和帧间特征的人数统计.pdfVIP

文嘉俊等：基于AdaBoost 和帧间特征的人数统 基于AdaBoost 和帧间特征的人数统计 1) 2) 3) 文嘉俊 徐勇 战荫伟 1) 2) 3) （哈尔滨工业大学深圳研究生院, 深圳 518055） （广东工业大学计算机学院， 广州 510006） 摘 要： 视频监控中出入口人数统计的难点在于人流密集时对每个人体的准确分割。针对此问题，本文通过学 习的方法得到人头检测的分类器，并在垂直拍摄图像中提取人头候选区域，以分离相互靠近的人体目标。进而根 据人头的运动特征剔除静止误检区域，根据误检目标检测频率低及其响应位置不连续的特征剔除动态误检区域。 最后提出一种简易可行的过线跟踪方案以完成计数。实验中对双向同时过线，搬货物过线和光照发生变化时过线 的视频进行测试，正确率能够达到95%以上。 关键词： AdaBoost 人头识别 帧差法 检测响应特征 目标跟踪 People counting based on AdaBoost and inter-frame features WEN Jiajun1), XU Yong2), ZHAN Yinwei3) 1) 2) (Department of Computer Science and Technology, Harbin Institute of Technology Shenzhen Graduate School, Shenzhen, 518055) 3) (Faculty of Computer, Guangdong University of Technology, Guangzhou, 510006) Abstract: The difficulty of entrance people counting in video surveillance lies in the correct segmentation of each person under crowded situation. For this problem, first, we learn a classifier to detect the head candidates so as to distinguish the people who walk closely under the overhead mounted camera. Furthermore, the static false detected candidates are removed according to the motion feature of the head, while the dynamic ones are eliminated due to their low detected frequency and discontinuous response positions. Finally, an easy and effective crossing tracking scheme is proposed for counting. Experiments are conducted under several situations, including bidirectional crossing simultaneously, crossing with items and crossing under lighting changes, which show that the counting rate can achieve above 95%. Keywords: AdaBoost, head recognition, inter-frame differencing, detection response features, object tracking