Identification And Analysis Of Abnormal Behavior Of In-car Video Monitoring

In recent years,with the rapid development of transportation and the improvement of residents’ quality of life,most people choose to travel by car for the convenience of movement.However,in recent years,there have been numerous incidents between passengers and drivers due to disputes,and even life threatening.Therefore,in the process of driving,in order to ensure personal safety,it is necessary to actively monitor the behaviors of the relevant persons in the vehicle in real time,identify and detect their movements,and provide early warning of abnormal behaviors.This article is mainly to intelligently identify and analyze the abnormal behavior in the video surveillance scene in the car.The abnormal behavior in the scene is divided into two types of threat and attack,and different methods are used to identify and study.The main tasks as follows:(1)For threatening abnormal behaviors in the car,including holding a knife,holding a stick,covering mouth,covering eyes,and pinching the neck,a neural network structure for real-time abnormal behavior detection is constructed based on deep learning.In order to meet the real-time nature of the detection system,this paper selects the end-to-end target detection algorithm YOLO to simply identify its behavior and objects,and initially get the character behavior and object categories.And according to the actual application of the project,the original YOLO network was improved.Secondly,in actual scenes,the behaviors of the characters have a certain degree of relevance to the objects around them.Considering this kind of problem,this article associates the behavior of the person identified in the scene with the interacting object.When the target feature is identified,these features are fused,and then the person feature and the interactive object feature are jointly calculated to obtain the specific behavior.Finally,the improved network structure was used to perform relevant experiments on the self-collected video data set in the car and the PASCAL VOC2012 behavior data set.The results of the study show that the neural network structure proposed in this paper for the identification of in-vehicle threatening abnormal behavior is increasing.While improving the detection speed,it also improves the recognition rate.(2)For the identification of attacking abnormal behavior in the car,this study is mainly realized by optical flow method.Because compared to normal scenes,sudden fights areusually accompanied by large,fast,and disordered movements,the participant’s characteristic point optical flow will change significantly,and the changes between consecutive frames are not regular.This paper first extracts the feature points of the target,and uses them to describe the moving objects.According to the analysis,the Harris corner method is used to extract the target features.Next,the pyramid Lucas-Kanade optical flow method is used to track the extracted feature points to form an optical flow field.Use the optical flow vector information in the optical flow field to extract the motion feature of the behavior,and compare the calculated motion feature value with a preset threshold to determine whether the behavior is abnormal.Finally,through the experimental test on the collected video data,the feasibility of the method for detecting abnormal behaviors in the car designed by this article is verified.