DSCam: uma plataforma hardware-software para operações de visão computacional

Abstract

The advent of integrated circuits with low power consumption and powerful processor capability, ally to many others news tecnologies, has became possible a new generation of systems with embedded inteligence, as smart-cameras. The smart-cameras are embedded vision systems that capture high-level descriptions of the scene and analyze it to extract important datas to an application in real time. However, the design of smart-camera is a challenging because on one hand video processing has insatiable demand for performance and power, and on other hand embedded systems place considerable constraints on the design. There is also the specific algoritms used in image processing, that will be adapted to processor architecture to obtain a best use of the resources. With the goal of facilitate this development grew the idea of Digital Smart Camera (DSCam). DSCam is an integrated solution for computer vision applications, associated with a development framework to help architects and programmers of the vision applications. Using the DSCam the programmers unaware the architecture details, working in a high-level language. In this work a comprehensive study of vision architectures is done and the DSCam is proposed, implemented, and tested. The results are satisfactory, with a very adaptable platform, an easy development interface and executes times near of similars works.