Research On FPGA-based Multi-moving Target Detection Technology

As a research hotspot in the field of image processing,the real-time detection technology of moving targets has been applied to many fields such as autonomous driving,machine vision,human-computer interaction and so on.However,in most cases,there are more than one moving target in these actual scenes,and the detection and processing time of these targets should be as real-time as possible.Therefore,it is very important to detect and mark multiple moving targets in a video image accurately after real-time detection.In addition,with the improvement of target extraction accuracy,the calculation amount of the algorithm tends to increase accordingly,resulting in poor real-time detection.Using a Field Programmable Gate Array(FPGA)with parallel processing capabilities to accelerate the target detection algorithm is an effective way to solve this problem.This paper first gives a brief introduction to the research on target detection based on FPGA at home and abroad,and points out some drawbacks and shortcomings.Then,in view of these existing problems,based on the characteristics of FPGA's parallelism and pipeline processing,the target detection and some other auxiliary algorithms that can be achieved on the FPGA are analyzed theoretically.Subsequently,the basic requirements and the overall structure of the multiple moving targets' detection system described in this paper are analyzed.A low-cost but cost-effective Spartan6 development board from Xilinx is selected,and a realtime multiple moving targets' detection system composed of image acquisition,image preprocessing,image storage,image target detection and image display units are built.In order to speed up the processing of the system and obtain good visual effect at the same time,the color video stream collected is divided into two streams: gray scale and color.Gray scale video stream is used for target detection,and color video stream is used for marking and displaying the detection result.The key components to ensure the real-time detection of the system are the image target detection unit and the image storage unit.The former consists of specially designed modules such as image difference,morphological processing,image segmentation,moving target position extraction,etc.The latter divides DDR3 into five storage areas with different first addresses,and stores image data streams in two formats using ping-pong operations.When implementing image format conversion,filtering,image difference,morphological processing,and Sobel detection,the parallel pipeline method of FPGA greatly accelerates the operation speed.Two methods are used to mark the detected moving targets,rectangular box marker based on image segmentation and edge marker based on edge detection,and their location information is extracted,which is mapped to a color video image.In this way,the function of real-time detection and marking of multiple moving targets in color video images under different conditions is realized.Finally,a series of experiments are designed and completed under different resolutions,frame rate,target motion speed,background change or not.Based on these experimental data,the performance of the system is analyzed,and several valuable conclusions for the design of real-time moving target detection system are drawn.