A Codesign Approach for a High Performance Vehicle Detector

摘要

A high performance architecture for an image processing application, a computer vision system for vehicle detection, is proposed in this work. The system aims to detect and count moving vehicles crossing a selected window from a video input captured by a traffic surveillance camera. The system works under a set of design constraints such as communication throughput, processing time and implementation cost on a real-time situation. The vehicle detector attains high performance on a low cost hardware/software platform, even using 2D FIR filtering, histogram thresholding and color space transformations, among others image enhancing techniques. This paper presents the development of the algorithms in a hardware/software DSP platform based on a field programmable gate array architecture, their implementation and the performance analysis on an application with real data.