Research And Implementation Of Pedestrian Abnormal Behavior Detection System Based On MIL

Intelligent monitoring system plays an important role in social security and criminal cases detection.However,most traditional intelligent monitoring systems cannot detect abnormal behaviors of pedestrians and record abnormal information automatically,which leads to the fact that it takes a long time for relevant personnel to retrieve abnormal behaviors of pedestrians after the abnormal situations.In this article we selected a deep MIL(multi-example learning)sorting model to detect 13 kinds of pedestrian abnormal behavior based on the characteristics of pedestrian abnormal behavior data.This model takes normal and abnormal video as packages,takes video fragments as examples in multi-example learning,so as to predict abnormal video fragments by using the idea of MIL.In order to get better training effect,sparse constraint and smoothness constraint were added to the model to locate the abnormal area more accurately.At the same time,a pedestrian abnormal behavior detection system was designed based on the model.The main work of this paper was as follows:1 We studied on the deep learning framework Keras,activation function,regularization methods,and traditional multi-example learning methods.2 Analyzed the functional,non-functional requirements and feasibility of the MIL pedestrian abnormal behavior detection system,determined the system as B/S(browser server structure)architecture according to the requirements,and designed the overall framework,functional structure,whole process of the system.Then on the basis of the system process the conceptual structure of database,logical structure and physical structure were designed.3 The C3D(3D convolution network)model was selected to pretreat the data set,extract the video characteristic data and fuse the extracted of sixth layer of full connection layer video feature data of C3 D model.4 After analyzing the multi-example learning model,the deep multi-example sorting model was finally selected to train and test the characteristic data,and a specific verification program was written to verify the trained model.5 Completed the pedestrian abnormal behavior detection system based on multi-example learning,set up the system server,configured necessary hardware and software environment,improved the functional modules,deployed the system to the local server,and tested the system from the two aspects of system's functionality and stability.Finally,the debugging of the pedestrian abnormal behavior detection system was successful.It is found that the system can well fulfill the system requirements through testing the pedestrian abnormal behavior detection system based on MIL.According to the testing results of specific video data,it can detect the time period of abnormal behaviors of pedestrians in video,and retrieve the corresponding historical abnormal information by using the time of occurrence of the abnormal events.