Research Of Moving Objects Detection System In Dynamic Background Based On FPGA

Video-based Moving Objects Detection is a hot research in computer vision area, related to image processing, pattern recognition and artificial intelligence and other fields. Accurate real-time moving object detection is the premise of the video tracking and analysis, which has great theoretical value and practical value. At present, the basic research of target detection algorithm has matured, but in complex dynamic background, because target detection algorithm is high computational complexity, the resulting is in poor real-time. With embedded hardware to achieve real-time dynamic background object detection is an effective way to solve the problem. This study has important practical value and market prospects.FPGA has high parallel efficiency, and satisfies the needs of the large number of data processing, which is very suitable for video image processing. Using FPGA as a hardware development platform, the paper realizes moving objects detection in dynamic background, based on the Institute in-depth research of moving object detection algorithm in complex background, adaptive background modeling and updating algorithms and multi-color space information fusion algorithm for many years, through the improvement of algorithm optimization. The paper includes the following sections:1. The paper introduces the importance of the embedded motion detection system based on FPGA,and the research status at home and abroad. FPGA technology development platform is also introduced, In the paper we mainly introduces the development board used in the subject, and the selection of the development board; the Quratus II integrated development environment and the application of FPGA-based system development processes are also detailed.2. The paper introduces basic knowledge and common method of target detection. And discusses in detail the key problems and difficulties of target detection in the dynamic context - the background modeling and updating methods. Subsequently, the article made a systematic analysis of the dynamic background modeling and updating the basic algorithm and found that one of the more sophisticated methods is based on Gaussian mixture model for background modeling and updating. Because a large amount of calculations in background modeling, so far, we have not seen the use of hardware circuits to implement the background modeling. In this paper, using the thinking of Gaussian mixture model the background modeling and approximate has been optimized, which efforts to reduce the computational complexity, makes it suitable for implementation on FPGA in order to ensure real-time, dynamic background to achieve a simple target detection.3. On FPGA, we design the hardware circuit to achieve the algorithm of background subtraction based on more color space, background modeling based on the Gaussian mixture model and ideological updating algorithm. First we get the background image frame, and then establish two Gaussian mixture model of the background frame. In order to reduce the amount of computation, In the design we only considered a parameter -mean , design a dual-port SDRAM controller and cache background model with off-chip SDRAM. Also achieve optimized edge detection algorithm in FPGA, and achieve the target contour calibration through edge detection, It is important identify targets.Background modeling and updating with hardware researched in this paper is a new direction, the related research is very unusual. The method of dynamic background modeling, which is effective for the rotation of fans, the screen flashes and other relatively simple dynamic background , and can meet real-time requirements. But for complex dynamic background modeling such as the woods, streets and so on, It is difficult to optimist its calculation, so the related use of FPGA needed further study.