基于位置和运动轨迹的老年人异常行为检测

摘要

For abnormal behavior detection of the elderly, some characteristics and rules of daily behavior can be gotten from the daily positon and motion trajectory. Through the sampling and quantization of trajectory, an efficient algorithm is proposed for abnormal detection based on the imroved HMM model and feedback sliding window. The trajectory landmark points and HMM initial value with the fuzzy C-means clustering is identified and the improved formula revaluation is given, then training and testing with feedback sliding window is carried on. Comparative experiments show that the method is effective for the abnormal detection in daily with position and trajectory and can ensure higher accuracy and lower false negative rate.%针对老年人异常行为检测,分析老年人日常行为特点及规律,对其日常运动轨迹进行采样和量化,并使用基于改进的HMM模型进行轨迹行为习惯建模,结合模糊C均值聚类算法提取轨迹标志点确定HMM初值,并通过改进的重估公式和反馈滑动窗口进行训练与检测.实验结果表明,运动轨迹可以作为老年人行为习惯的一个重要描述形式,利用位置和轨迹对家庭环境下的异常行为检测是有效的,且能保证较高的准确率和较低的漏检率.