基于轨迹词包模型的老人跌倒检测  

Study of Fall Detection for Elder People Using the Bags of Trajectory Words Model

作　　者：闻帆[1,2] 屈桢深[2] 闫纪红[1] 

机构地区：[1]哈尔滨工业大学机电学院,黑龙江哈尔滨150001 [2]哈尔滨工业大学空间控制与惯性技术研究中心,黑龙江哈尔滨150001

出　　处：《自动化技术与应用》2015年第2期60-67,共8页Techniques of Automation and Applications

基　　金：国家自然科学基金(61375046);中央高校基本科研业务费专项资金资助(HIT.NSRIF.2010074)

摘　　要：针对异常行为检测受到光照变化、目标遮挡和计算复杂度高等因素的影响而导致检测效果不理想的问题,本文提出一种基于时空兴趣点和轨迹词包模型的异常行为检测算法。首先,利用时空兴趣提取目标的特征点信息;其次,利用稀疏光流法对特征点进行跟踪,获取目标的运动轨迹。然后,利用Meanshift聚类算法对轨迹进行聚类并构建轨迹词包模型。最后,利用SVM完成异常行为的判别。算法在不同视频数据库上进行了验证,并取得了93.3%的准确率。通过与以往的实验结果的比较,算法在异常行为检测方面具有较好的实时性、准确性和可靠性。An abnormal behavior detection algorithm based on space-time interest points and bags of trajectory words model is proposed to solve the problem of poor performance in the course of behavior analysis, it is disturbed by light changes,high computational complexity, target occlusion and other factors. At first, targets feature information is extracted by using space-time interest point method. Secondly, the feature points are tracked by a sparse optical flow method, and obtained target＇s trajectories information. Thirdly, using Meanshift clustering algorithm to cluster trajectories and build a bags of trajectory words model. Finally, SVM is adopted to complete the task of abnormal behavior detection.The algorithm is tested on different video datasets, and makes a 93.3 percent accuracy rate. Comparing with previous experimental results in behavior detection, the algorithm has a better accuracy and reliability and a fast computing speed.

关 键 词：轨迹词包模型 跌倒检测 异常行为 时空兴趣点 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]