Research On Dectecting The Prisoners' Crossing-Border In Jail

With the advance of computer intelligence,security system posseses the potential of the intelligent trendcy.Its development and applications exempt people from staring at surveillance videos for a long time,meanwhile avoiding visual fatigue caused by the long-time work.This can eradicate false warnings in surveillance regions and prevent from the crimes and accidents.In addition,security systems can take full advantage of the computer vision technology for public opinion monitoring,intelligent transportation,to name a few.Henceforth,it is the most important task how to utilize the visual techonology to process surveillance videos in an effective yet realtime way,especially for mornitoring prisoner videos.To address this issue,due to the constant background of monitor regions,this paper proposes a fast robust PCA algorithm(Fast RPCA,FaRPCA)to real-time monitor prisoners by effectively learning the mornitoring background and extracting specific pedestrians in the foreground.Moreover,we further incorporates various traditional methods to design a cross-border detection system in prison,which distinguishes between warders and prisoners and avoids false warnings.Our main contributions are as follows:(1)We review three classical foreground extraction methods,and detail the basic principles and pros and cons of Gaussian background modeling,RPCA and Go Dec,respectively.(2)We propose an efficient foreground extraction method termed FaRPCA.Experiments on six benchmarks show that FaRPCA outperforms RPCA and Go Dec in terms of foreground detection efficiency and effacy.(3)We design a cross-border detection system in prison.By integrating FaRPCA foreground detection algorithm,Canny edge detection,Hough line detection,color recognition and focus detection method,this system can monitor prisoners real-time.