Research On Detection Of Abnormal Behavior And Track Strategy In Monitoring Environment

With the rapid development and popularization of the monitoring system,the coverage of the camera has been expanding,and the monitoring video data has witnessed an explosive growth.The emergence of big data technology can well solve the user's demand for monitoring data storage,but it cannot meet the growing demand for monitoring human abnormal behaviors in the monitoring environment.Therefore,how to use video data to analyze human behaviors and establish models to detect human abnormal behaviors in the monitoring environment has become a research hotspot.The analysis of human behavior in the video usually first extracts human behavior features,and then makes modeling analysis of human abnormal behaviors based on human behavior features.However,most studies only focus on the single feature of human like trajectory,attitude or speed,ignoring the relationship between them,which provides a new direction for the design and implementation of abnormal behavior detection algorithm.At the same time,after making abnormal behavior,it needs to track the human.This human is likely to leave the current camera.Therefore,cross-camera tracking of the human is necessary.The results of human behavior analysis and human features can effectively improve the speed of cross-camera tracking.This thesis is mainly to analyze human in the monitoring environment and analyze the dynamic behavior features of human to realize the intelligence of the monitoring system.The research content is mainly divided into three aspects: collection and preservation of human behavior features,the construction of abnormal behavior detection model,and the research of cross-camera tracking algorithm,as follows:1)In order to obtain the static behavior features of the human,the detection and segmentation of the human in the picture,the detection of the key points of the human skeleton and the detection of the physical coordinates of the human are firstly realized.The tracking of human in a single camera is realized by using the characterization features and technology of person re-identification.Based on this,a dynamic behavior feature acquisition algorithm of human is proposed to collect the motion track,attitude sequence and speed sequence of human in the video.Based on the features of human behavior,a prototype of intelligent monitoring system based on the characteristics of human behavior is designed.2)Based on the dynamic behavioral features of human,the relationships between each feature and abnormal behavior are analyzed.And the correlations between different features are also analyzed.The clustering algorithm is used to model the movement trajectory of human,and the probability model is used to subdivide the clustering based on attitude sequence and speed sequence.Based on the model,a multiple features based abnormal human behavior detection algorithm is proposed to reduce false alarm rate while realizing the fine-grained detection of abnormal human behavior.Finally,the data set is used to verify the performance of the algorithm.3)Based on the clusters,the monitoring range of camera is partitioned by clustering algorithm.Based on this,the topology of camera is constructed by using probability model.An algorithm associated with space and time for cross-camera human tracking is proposed,the improved person re-identification technology is used to accelerate the matching process.And the topology of camera and the speed of human are used to accelerate the tracking of human from the angle of space and time.Finally,the effectiveness of the algorithm is verified by simulation.