Multi-Camera Computational Video Architecture

Abstract

Image sensors have become ubiquitous. Adding extra image sensors to a vision system does not need to be costly. Processing all this information however, is non-trivial. The goal of this dissertation is to investigate a multi camera architecture especially designed for processing multiple image sensors for single chip implementation. This research aims to demonstrate and formalize better understanding on the design aspects and trade-o s. This will require the design, realization and implementation of dedicated hardware structures and architectures on a Field Programmable Gate Array (FPGA). A real-time stereo vision system is selected as use case for this dissertation. Instead of using one camera pair, the goal is to use multiple camera pairs. By using the cooperative results to obtain a good and robust depth information instead of pursuing the best results for each camera pair separately. By providing an architecture for multi camera systems and their collaboration, we have shown that it is possible to implement them into a single chip implementation.To illustrate this approach, a depth camera has been developed which is working according to this principle. It is a true single chip implementation.