For Real Time Application Of Video Detection Algorithm Research And Fpga Implementation

Video object detection, tracking and recognition are always the hot spot in intelligent surveillance, pattern recognition and video searching study field. These technologies which closely related to daily life have been widely applied. Detection is the first step before other high level processing such as tracking and recognition in this field. The precision and efficiency of detection results have a directly influence on the following steps. The detection methods based on detect elements can be divided into two classes:first, those used pixel as the basic elements can be called detection in pixel domain; and second those used DC parameters, motion vectors as the basic elements during video decoding can be called detection in compressed domain. I focus on real time video detection doing some study on these two domains respectively.In compressed domain, this paper based on H.264protocol discusses following topics: first, picking up motion vectors from H.264decode stream; second, building a outstanding vector field from raw motion vectors; third, doing detection algorithm based on a enhanced clustering algorithm. Simulation results show that this enhanced algorithm has a higher detection speed compared with the original one. It can also effectively eliminate optic noise influence. In addition. It can detect the object which is temporarily static.The advantage of compressed domain is that it start doing detection when the decoding is not over, moreover, the processing data is less compared with the pixel domain, both of which can increase the efficiency of detection. But because the basic detection element is4X4motion vectors, its precision has limitation compared with the pixel domain.Pixel domain study has a long history. Its theory is more strictly. But nowadays, high definition video is applied more and more widely. It means magnanimity pixel data. The traditional pixel detection can hardly both keep high precision and real time requirements. This paper raises an enhanced pixel domain detection based on change-detection-based approaches. It builds a reliable background buffer based on video input and gets the object field based on current frame, background buffer and previous frame. These approaches are high efficiency compared with the watershed ways. And it is suitable to be realized by hardware because its processing can be parallel and pipeline.In the view of realization, object detection is always finished by software with other high level processing because of considering system cost and the complex application environment. Nowadays, using FPGA to do detection realization can provide a well solution for real-time application. Compared with software realization, using FPGA will add additional cost, but as a programmable chip, it can do optimization following dedicated algorithm. It can hugely increase the processing speed because of the hardware advantages such as parallel and pipeline.This paper based on FPGA development platform, design a detection system based on the above pixel domain detection approach. It use digital sensor as input, screen as output, SDRAM as solving buffer. And in FPGA, I design the sensor configuration module, pixel synchronization module, pixel transformation module, DMA module, SDRAM controller module, VGA controller module and Detect module. This system works well and meets the real-time detect requirements.