Moving Object Detection System Based On FPGA

Moving object detection in image sequences is an important research direction of computer vision, and it’s widely used in safety monitoring, intelligent transportation and so on. With efficient parallel processing capability, Field Programmable Gate Array (FPGA) provides an excellent platform to real-time implementation of complex image processing algorithms, the image processing system based on FPGA has broad prospects for development. This paper designs a moving object detection system based on FPGA, and it is flexible for later changing of algorithm, collection and display equipments.In this paper, by analyzing the common moving object detection algorithms, a background subtraction method which is based on adaptive Gaussian Mixture Models (GMM) is adopted. Hardware complexity of GMM is analyzed firstly, then this paper adopts fixed-point processing and add scale-down scheme for GMM algorithm in OpenCV. We verify the performance of the modified algorithm by Matlab simulation, achieving the balance between algorithm performance and the implementability on FPGA.The moving object detection system designed in this paper collects real-time image by high-speed CMOS image sensor, this paper realizes real-time Bilinear interpolation by the line-buffer structure, and then transforms Bayer image to RGB image. This paper realizes frame rate conversion from camera collection data to DVI display data by using a QDR Ⅱ+SRAM and FPGA drives DVI display chip-TFP410, realizing1024x76860fps DVI display on the general LCD monitor. The system also includes a PC controller scheme, realizing that the PC configure I2C device in FPGA board through the UART serial port, I2C bus and protocol conversion logic.In the processing of realizing the moving object detection algorithms, this paper designs multi-level pipeline and parallel structure. This paper uses a QDR Ⅱ+SRAM to store one frame’s Gaussian parameters. The FPGA system operates at65MHz, and spends about6ms completing the detection of one frame. Compared with the software processing time,230ms per frame, the detection speed is greatly improved. The system highly meets the real-time requirement of30fps, and this paper uses FPGA to achieve the algorithm hardware acceleration very well.This paper successfully realizes moving object detection of1024×76830fps video on Altera Stratix IV development platform. The results of the experiment show that the system has good detection effect and anti-jamming capability.