Video Surveillance System With Automatic Motion Detection In Multi-zone

Automatic motion detection can reduce the human cost of video surveillance and improve efficiency,it is also the fundament of object extraction,tracking and recognition.This technique can be extensively used in security monitoring,traffic safety and production safety systems.In this work,efforts were made to establish the background model which is resistance to the variation of illumination.The video surveillance system was also realized by a FPGA based platform.To reduce the false detection rate caused by the shifty intensity and angle of illumination,we proposed an algorithm based on improved LBP pattern,which is a powerful texture description and comparison mechanism and resistance to the variations of illumination.In addition,we proposed an adaptive background model updating algorithm which updates the background according to the result of frame difference.In this way,one can avoid the misidentification of moving object and background.We also proposed an algorithm which can detect moving object in independent zones simultaneously.Besides,this algorithm can resist the interferences from the substances,such as the quivering leaves and water surface.Above algorithms have been implemented on a FPGA based platform in the form of hardware logic.The pipelines,parallel structures and data storage structures are designed accordingly.Several functions have been realized in hardware,such as the window extraction,mean value filter and LBP pattern calculation,background refreshing,frame difference and background subtraction.Based on these functions,we can realize a real time moving object detecting system.This video process system can be divided into four modules:moving object detection module,video input and output module,frame buffer module and surveillance area control module.Video I/O module includes video interface chips,video path and PicoBlaze MCU.This module can also superimpose the framework of target areas to the video.Frame buffer module includes a DDR chip,a MPMC IP core and read/write control state machines.Surveillance area control module provides the user with an interface which can be used to control the location and size of the target area.This system can realize real time detection in up to 4 areas independently.At the end,we used Chipscope logic analyzer to verify the correctness of the timing of each module.A verification experiment was designed and the ROC curves of both of the LBP and grayscale-based algorithm ware drawn and compared.One find that the algorithm based on LBP is more robust to the change of illumination.Experiment results demonstrate that this system can execute moving object detecting and background refreshing algorithm correctly.It supports a VGA video processing up to 1024×768/60fps,and can output a clear and stable video signal.