The Design And Implementation Of Video-based Abnormal Behavior In Crowds Detection System For Public Places

With the improvement of public safety awareness and the reduction of the manufacturing cost of video monitoring equipment,video monitoring system has been set up in more and more places.Among them,the anomaly detection in crowds system for public places is particularly important.The high crowd density in this scenario leads to the increase of the probability of abnormal events and the expansion of its influence scope.However,the existing anomaly detection systems are mostly designed for specific places or abnormal behaviors,lacking the universal detection system for public places.Therefore,this study combined with deep learning technology to study the relevant fields of abnormal behavior detection in crowds,designed and implemented an universal abnormal behavior in crowds detection system.In this study,we take the abnormal behavior in crowds detection as the core part.Anomaly detection in surveillance videos is a challenge problem due to the ambiguous definition of ‘Anomaly' and the complexity of abnormal events.We propose a novel spatio-temporal generative model to solve this problem by constructing a two-stream generative model,which perform video frames reconstruction and future frame prediction at the same time to fully utilize the information of input video.For enhancing its ability of capturing spatio-temporal features,which represent the apparance and motion in video sequences,of normal video patterns.we firstly intergrate ConvLSTM with this encoder-decoder format generative model,then construct a discriminator and train them in an adversatial way.After traing on normal videos,our generative model can be independently used as anomaly detector by calculating reconstruction and prediction error on its input video sequences.Experiments are performed on three public datasets for examining the effectiveness of our model,which show our method is competitive with these state-of-the-art approaches.We further visualizing the location of detected anomaly in frames for model interpretation.In addition,considering that crowd congestion often occurs in public places and the abnormal detection model lacks the ability to foresee the occurrence of abnormal behaviors,we designed and used a crowd density estimation model to obtain crowd density information in video.Based on the usage of this model to get the the crowd density distribution and estimate the number of people,we can alarm security personnel to global crowd congestion and local crowds gathered phenomenon,help them dispersing crowd and discovering abnormal events to avoid their occurance or reduce impacts.We design the model based on the latest research and verify its validity with a dataset.At the same time,based on the generated crowd density distribution map,we also carried out a visual display of the density distribution to explain the working principle of the model.Finally,based on the above research results,this study carries out the demand analysis of abnormal behavior in crowds detection system,the design and implementation of the prototype system,the function and performance test of the system.The test results show that this system is adapted to different scenes,and can effectively and real-time realize the detection and location function of abnormal behavior in crowds and crowd density abnormality.In addition,the crowd density estimation module added in the system can assist the detection and prediction of abnormal behavior in crowds and enhance the practicability of the system.