基于统计模型的异常姿态检测

Statistical Model-based Abnormal Behavior Detection

随着视频监控系统在各领域越来越广泛的应用,传统的人工标定视频中异行为的方法,已不能满足海量视频数据分析的需求。通过分析目标的姿态,能够检测出瞬时或短周期目标动作的异常。这里提出一种基于姿态码本和姿态直方图的异常姿态检测方法。该方法不依赖场景的先验知识,受视点变化的影响较小,模型能无监督实时更新以适应时变环境,能有效地检测目标的姿态异常。实验表明该方法的有效性和鲁棒性,具有一定的应用价值。

calibration of abnormal behavior in video now could not meet the demand for analysis of massive video data.By analyzing the target’s gesture,the anomaly of target’s instantaneous motion could be detected.Based on gesture codebook and gesture histogram,an abnormal gesture detection method is proposed,and this method relies on no priori knowledge of the scenario and is less influenced by the change of viewpoint,the model could be updated in real time and without any supervision,thus to adapt itself to the time-varying environment,and effectively detect gesture anomaly of the target.The experiments indicate the availability,robustness and the application value of this method.