Research And Realization Of Video Moving Object Detection System Based On FPGA

Moving objects detection has extensively been applied in both military and civil areas, and has very important theoretical significance and practical value for rapid and accurate moving objects detection. Many progresses have been made in the study of moving object detection algorithms; however, it is uncommon to realize these algorithms using hardware which is still a research hotspot at present. FPGA (Field Programmable Gate Array) can be used to satisfy the requirement of high speed, integration and reliability of video processing system as a hardware platform, which is suitable for the application of video and image procession. On the foundation of extensive researches in moving object detection algorithms these years, we managed to realize these algorithms by FPGA. Up to now, gray-level fluctuation and gray-level shift and information fusion of multiple color spaces have been realized by FPGA. The main points are as follows:1. Video moving objects detection system has been established based on FPGA, and captured video can be displayed in real-time. The system consists of five modules: video capture, video preprocessing, video moving object detection, video display and warning system. Firstly, in the video capture module, analog video signal captured by camera is sent to video decoder ADV7180, which converts the analog video to 8 bits serial digital video data YCbCr 4:2:2 compatible with CCIR656 standard via making proper configuration for the decoder registers by the I2C bus configuration module designed in FPGA; Then, in the video preprocessing module, digital video date YCbCr 4:2:2 is decoded in ITU-R656 standard and is converted from YCbCr color space to RGB color space in FPGA to get RGB tricolor; After that, video data gray processing is performed, by which we get gray value form RGB value for next processing step; All of these were done in FPGA by Verilog hardware description language programming; At last, processed video data is converted to analog signal by digital-analog conversion chip ADV7123, and is displayed on CRT display. Thus, video capture and video preprocessing are realized, and captured video is displayed in real-time.2. Background differencing detection algorithm has been successfully realized in FPGA, and moving objects in video are detected, tracked and alarmed in real-time. After gray processing, current frame images and background images selected from video are sent to moving object detection module at the same time after buffer in SDRAM. In this module, background difference algorithm is used to find out moving objects in current images, and the moving objects were marked by black pixel blocks. When the objects move, the black pixel blocks closely follow them, thus the objects are detected and tracked in real-time. At the same time, FPGA sends out alarm signal to audio decoder, and gives warning sound through loudspeaker. Detection experiment shows that moving objects in video are detected and tracked in high speed by the system.3. Improvement of background differencing algorithm is made in FPGA. Firstly, to get accurately detecting and positioning of moving objects, information fusion of multiple color space is added to the system, which uses many complementary components in multiple color space to determine moving objects, that means we use not only the gray value in background differencing algorithm, but also other color space components such as U, V, R, G, B, H, S, V to make background difference together, and the result shows that the object region are marked of more black pixel blocks, the shape of moving objects are embodied more accurately. Further more, gray-level fluctuation and gray-level shift algorithms are also added to the system, which can remove the excessive noise points after background differencing algorithm, and also eliminate pixel gray-level shift caused by micro vibration, air disturbance and so on. The black pixel block mark in object region is more apparently, while the noise points are mostly disappeared.This system can be used to detect moving objects under static background and also light slightly changed background. Gaussian mixture module will be used to detect and track moving object under dynamic background such as wind-blown trees, hassock and water wave in next step.