室内视频监控下基于混合算法的人体异常行为检测和识别方法

HUMAN ABNORMAL BEHAVIOR DETECTION AND RECOGNITION BASED ON HYBRID ALGORITHM IN INDOOR VIDEO SURVEILLANCE

针对目前室内人体异常行为检测和识别中照明变化、遮挡、相机移动和背景等因素对检测准确性的影响,提出一种多技术混合跟踪方法。该方法基于连续自适应均值漂移(CAMS),引入校正背景权重直方图(CBWH)和无味粒子滤波(UPF)技术处理遮挡和相似颜色对象的干扰。采用基于稀疏表达的检测方式从多种场景对目标对象的异常行为进行检测和识别,并利用均方误差统计量评估所提方法的性能。同时在公开数据集UMN上进行仿真验证。实验结果表明,该方法在不同场景中有障碍物遮挡或是具有相似颜色的其他对象情况下都能准确检测和识别目标对象。此外,该技术还可能进一步改善复杂场景下多摄像机中目标对象的跟踪性能。

Aiming at the influence of illumination variations,occlusions,camera movements,and background clutters on monitoring accuracy in the detection and recognition of human abnormal behavior,a hybrid multi-technique tracking method was proposed. Based on continuous adaptive mean shift(CAMS),we introduced corrected background weight histogram(CBWH) and unscented particle filter(UPF) to deal with the interference of shading and similar color objects. The sparse expression detection method was utilized to identify the abnormal behavior of the target object in various scenarios. The performance of the proposed method was evaluated by using the mean square error(MSE),and the simulation was carried out on the open data set UMN. The experimental results show that the method can accurately detect and recognize the target object under the condition of obstacle occlusion or similar color object in different scenes. In addition,this technique may further improve the tracking performance of target objects in multiple cameras and complex scenes.