Distributed Object Detection And Its Parallel Implementation On FPGA

The application of object detection technology is becoming more and more various.With the development of science and technology,this technology has permeated through various kinds of leading-edge areas all over the world.As one of the most important features of the image,the edge information of the image is a key technology in target detection researches,it has been utilized into the segmentation,detection and recognition of images.Many studies have the edge information of the image with the image texture,color and other characteristics applied to the target detection to obtain a better robustness to illumination and pose changes.But the owing to the large amount of calculation,the traditional software is difficult to meet the real-time request.To solve the problem,the main research of this paper is divided into the following points:(1)Sobel is one of the most commonly used gradient calculation of edge detection operator in practice,to solve the problem aroused by the algorithm's using the global threshold and the illumination changes,This paper presents a distributed Sobel algorithm for image edge detection.According to the characteristics of human visual perception model,this paper build a threshold function which is adaptive to the images' partial gray feature to solve the problem caused by overall threshold in order to improve its performance,we construct the gradient kernel function to solve the difficult gradient detection of low gradient part of the images caused by the reasons such as the illumination changes.(2)Aiming at solving the problem of high real-time requirement and large amount of data for calculating in Sobel,this paper proposes a parallel structure of distributed Sobel operator in FPGA.We set up the system structure and finish the simulation In the simulink/XSG,and build the image acquisition and display platform through hardware programming language in FPGA,the adaptive algorithm and the gradient kernel function are realized by generating MIF file into the Block RAM to initialize it,by this way we decrease the system delay and improve real-time performance.(3)Aiming at the meeting the need that object detection operation often need timely access to the outline information of face,hands and other exposed parts of the human,this paper designs a method to extract the outline of skin parts combining the gradient information and the color space,and gives a FPGA parallel structure based onYCbCr color space and edge detection algorithm.This paper uses 3×3 templates to calculate the partial gray feature value and the partial area threshold,and improves the system robustness to situation like illumination changes,and by using the gradient kernel function,we improve the detection effect of low gradient section.In this paper,a fast skin color outline detection method is proposed based on the combination of the algorithm and skin color detection.Algorithm is built and simulated on the XSG and mapped to the Altera Cyclone FPGA platform,the image data stream from the OV7725 acquisition of 640 x 480@30FPS image to meet the real-time requirements.