Systolic Video Stream Object Detector Using FPGA-E

Dr.Shefa A Dawwd,Ula T Salim

doi:10.33899/rengj.2014.89977

Abstract

Object detection is important operation used in multiple applications such as computer vision, image and video processing, security, artificial intelligent, and several other areas.However, in these applications, it is not easy to realize real-time frame rates and fast invariant detecting function under changing object states such as position and size using software implementations. So that to solves these problems and speed up the highly intensive calculation required, In this paper simple and efficient template matching algorithm architecture of a video streaming application for object detection is proposed,it is based on using Sum of Absolute Differences (SAD)withPyramid Sum of Absolute Differences (PSAD) as similarity measures and a systolic array design using sliding window operation, where each video frame is divided into slides and feeds through the window by using a suitable first in first out(FIFO) buffers instead of the sliding window across the video frame. The implementation operation is done by using combination of software and hardware co-design that is based by using pipelining technique, data recirculation , and single instruction multiple data (SIMD) operations. The results for both SAD and PSAD algorithms showed the best match can be found at the template (window) size is 19×19 bits/pixel and with accuracy detectionrate of100%. Keywords: FIFO, FPGA, Object detection, Pipeline, PSAD, SAD,Sliding window, Systolic array, Template matching, Video stream.

Highlights

Video streaming can be defined as the process of applying some operations on video frames, streamed on a system picture by picture
In 2012 N., Dawoud, et-al [6] proposed a fast template matching technique based on Optimized similarity measurement metrics namely: Sum of Absolute Difference (OSAD) and Sum of Square Difference (SSD) to overcome the drawback of Normalized Cross correlation (NCC). This similarity measure application on face localization and with the other similarity measurements .the results show the highest performance of OSAD compared with other measurements and the improvement of OSSD comparing with SSD as well
The results show that FPGAs can achieve speedup of up to 11x and 57x compared to Graphics Processing Units (GPU) and multicores, respectively, while using orders of magnitude less energy

Summary

Introduction

Video streaming can be defined as the process of applying some operations on video frames, streamed on a system picture by picture. The frames are usually transmitted pixel by pixel and can be processed on a pixel by pixel basis. Most of the video streaming systems are built on a chain structure made up of a set of blocks. The first block on the chain is in charge of capturing the video frames, while the last block usually deals with the rendering of the frames. Between the first and the last modules, several operational blocks can be used according to the implemented algorithm [1]. There are different processes performed by Video processing applications such as video image enhancement [1], and object detection [2]

Methods

Results

Conclusion