VORTECS: VLSI Object Recognition Triainable Embedded CMOS System
logo home project info schematics programs credits
send e-mail
>> Abstract

The purpose of this project is to create a system for three-dimensional machine vision. A novel method is proposed to generate curves and vector lines in two- and three-dimensional space rather than the use of a large depth map. Thus, scene complexity dictates size of the stored object, and resolution is very high. Without the use of a controlled environment, features are stereoscopically matched in extremely small search space for an output of three-dimensional faces and curved surfaces. By implementation of this onto a Xilinx Spartan 3 FPGA in VHDL, existing and newly created algorithms are modified for real-time performance; many capable of running at the same speed as the incoming video, with a system clock of 50MHz. Digital cameras are modified for a high resolution color input. Raw data and sync signals are taken directly from the CCD processing chip within each camera. This Bayer coded data is interpolated into RGB and YUV, averaged with recent data, then presented to the filtering algorithms, which find object edges, represent the edges as curves, find the possible match space, stereoscopically generate the three-dimensional output, and present to a multilayer neural network. This network is implemented in two mediums: a MATLAB computer simulation and design of an analog VLSI CMOS 0.35um chip using the TSMC process. Design uses FGMOSFETs as synapses to store and update weights. Data representation by angles and distances as well as connections allows for built-in invariance to rotation and translation in three-dimensional space.

>> Project Report and Movies


Adobe PDF - Project Report (42 pages)
Size: 3.1MB
Click here to Open (right-click, "save target as" to download)

Adobe PDF - Robot Vision Algorithms Report
Size: 0.25MB [needed]
Click here to Open (right-click, "save target as" to download)

Macromedia Flash - Introduction
Size: 0.22MB [needed]
Click here to Open (right-click, "save target as" to download)
Copyright © Malcolm Stagg 2005. All Rights Reserved.
Website: http://www.alumni.ca/~stag5m0. E-mail: malcolmst@shaw.ca.