室内场景的异常行为检测与识别技术研究

发布时间：2018-06-04 00:29

  本文选题：智能监控系统 + Hu矩　； 参考：《西南科技大学》2016年硕士论文

【摘要】：智能监控系统因其全天候、无间断、低误报实时监控的优点而广受关注,其中的目标检测、目标跟踪和行为识别等关键技术是学者们研究的热点。针对室内固定场景,深入研究了目标检测、目标跟踪与行为识别技术,并分别针对存在的问题做出了改进。目标检测部分,将基于Vi Be的背景差法与帧差法融合,通过判断是否发生光照变化来选择对当前帧图像进行目标检测的方法,解决了Vi Be算法在光照变化的情况下检测到的运动目标不准确的问题。目标跟踪部分,利用卡尔曼滤波器的运动估计来改进Camshift目标跟踪算法,通过Bhattacharyya距离和遮挡率来判断目标是否被遮挡以及被遮挡的程度,能够有效解决目标在发生遮挡时跟踪不稳定的问题。异常行为识别部分,提出一种改进的基于模板匹配的人体目标异常行为识别算法,将改进的Hu不变矩和图像运动特征结合组成行为特征向量,采用Hausdorff距离计算待测行为特征向量与模板之间的相似性,并通过相应的阈值判定待测行为是否属于异常行为。实验结果表明改进目标检测、跟踪和识别算法均可行有效,并且提高了异常行为的识别率。
[Abstract]:Intelligent monitoring system has attracted much attention because of its advantages of all-weather, uninterrupted, low-false alarm real-time monitoring. The key technologies such as target detection, target tracking and behavior recognition are the research focus of scholars. The techniques of target detection, target tracking and behavior recognition are deeply studied for indoor fixed scenes, and the existing problems are improved respectively. In the object detection part, the background difference method based on Vi be and the frame difference method are fused to select the target detection method for the current frame image by judging whether the illumination changes or not. The problem of inaccurate moving target detected by Vi be algorithm is solved. In the part of target tracking, the motion estimation of Kalman filter is used to improve the Camshift target tracking algorithm, and the Bhattacharyya distance and occlusion rate are used to judge whether the target is occluded and the degree of occlusion. It can effectively solve the problem of tracking instability when occlusion occurs. In the part of abnormal behavior recognition, an improved algorithm based on template matching is proposed. The improved Hu invariant moment and image motion feature are combined to form the behavior feature vector. Hausdorff distance is used to calculate the similarity between the feature vector and the template of the behavior to be tested, and the corresponding threshold value is used to determine whether the behavior under test belongs to abnormal behavior. The experimental results show that the improved target detection, tracking and recognition algorithms are effective and the recognition rate of abnormal behavior is improved.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
，

本文编号：1974959