Design And Implementation Of Video Abnormal Behavior Detection System Based On Hadoop

In recent years,with the rapid development of society and economy also with the gradual popularization of urban and rural surveillance equipment,various anomalies and criminal behaviors under video surveillance can be recorded in video files.However,at this stage,supervision of video still depends on manual supervision,which will not only cause a waste of manpower cost,but also will inevitably cause mistakes and cannot prevent various crimes and dangerous behaviors in time because the video supervision requires an continuous mode throughout the day.Therefore,how to effectively obtain abnormal events in surveillance video has become an urgent problem.This article studies the extraction of abnormal behavior events based on people or objects from massive video data,and judges the abnormal events in video surveillance.First of all,this article introduces the relevant technical theories required in detail during the system establishment process,including FFmpeg,fuzzy set theory,YOLO algorithm,the distributed processing system Map Reduce in the Hadoop framework and the distributed file management system HDFS.Secondly,we propose Hadoop-based video abnormal behavior detection algorithms,including CAE-based and fuzzy theory-based abnormal behavior detection algorithms.Finally,we complete the design and implementation of various functional modules based on the characteristics of video abnormal behavior events,mainly include a streaming media server module which stores video data and forwards the video data in real time,a Hadoop framework processing the video data module and a video-based abnormal behavior detection algorithm module.In the video-based abnormal behavior detection algorithm module of this article,we use two methods to detect abnormal behavior.For the first method,we introduce an unsupervised feature learning framework based on object-centric convolutional autoencoders to encode motion and appearance information.Then,we propose a supervised classification method based on training sample clustering as normal clustering.Finally,a one-to-many anomaly event classifier is used to separate each normal cluster from other clusters.During the inference process,if the highest classification score assigned by a static classifier is negative,the object will be marked as abnormal.For the second method,we first use a method based on fuzzy theory to establish an abnormal event detection model,then establish a fuzzy discriminant formula based on the model attribute elements,and finally use the PSO algorithm to calculate the weight parameters in the discriminant formula.At last,this paper analyzes and tests the system functions using surveillance video data set up in schools.Experimental results show that the system can effectively detect abnormal events in surveillance video.