Recognition And Research About Abnormal Behavior Of Human Based On Video

With the continuous development of modern video capture technology and image acquisition hardware,people can obtain a large amount of video data,but how to collect and obtain valuable information from massive data,filter and reject invalid information becomes the key to computer visual video processing.Video anomaly behavior detection is one of the important research directions.The main research content is to analyze and process the abnormal behavior in the video,extract the data characteristics of the abnormal behavior characteristics and classify the abnormal behavior.In recent years,the deep development of deep learning algorithms in the field of computer vision has provided a feasible method for accurate and efficient detection of abnormal behavior in video.This thesis mainly uses deep learning methods to detect and analyze complex abnormal behaviors(such as violent behaviors)in video,and uses traditional computer vision algorithms to detect simple and explicit abnormal behaviors(such as running and cross-line detection).Different algorithms perform identification tasks efficiently and accurately.This thesis focuses on the abnormal behavior from the video,and analyzes and verifies the practical significance and application value of the algorithm in the detention center and prison security.During the work period,relevant construction opinions were put forward in the draft opinions on the construction of“Wisdom prison”and“Science and technology diligent”.According to the characteristics of complex anomalous behavior,this thesis firstly uses the Darknetl9 network architecture to design the CNN network is designed to automatically generate abnormal behavior data features.The LSTM network is better to deal with the continuous processing of motion in video.We proposed a target tracking algorithm based on Kernelized correlation filter algorithm and Kalman filter fusion.The artificial features detect the abnormal behavior in the video.Using the moving target detection algorithm,the abnormal behavior detection methods such as running and cross-line detection are designed.According to the algorithm,the training and testing were carried out in the abnormal behavior detection data set.The experimental results show that the algorithm has a great improvement in detection performance-The detection rate of violent behavior detection reaches 87%,The abnormal running detection and identification,the cross-line detection and identification method are simple,and the real-time performance is good,the target tracking effect is good,and the target could still follow the target after the target is occluded.Good,FPS can reach 63 frames/s.Finally,the module design and implementation of the system is completed,and the input video scene can be detected,and the alarm can be started in time when an abnormal behavior occurs.