Date of Award


Publication Type

Master Thesis

Degree Name



Electrical and Computer Engineering

First Advisor

Jonathan Wu


Applied sciences



Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.


The main motivation of the thesis is to develop a fully integrated, modular, small baseline (<=3cm), low cost (<=CAD$600), real-time miniaturized embedded stereo-vision system which fits within 5x5cm and consumes very low power (700mA@3.3V). The system consists of two small profile cameras and a dualcore embedded media processor, running at 600MHz per core. The stereo-matching engine performs sub-sampling, rectification, pre-processing using census transform, correlation-based Sum of Hamming Distance matching using three levels of recursion, LRC check and post-processing. The novel post processing algorithm removes outliers due to low-texture regions and depth-discontinuities. A quantitative performance of the post processing algorithm is presented which shows that for all regions, it has an average percentage improvement of 13.61% (based on 2006 Middlebury dataset). To further enhance the performance of the system, optimization steps are employed to achieve a speed of around 10fps for disparity maps in MESVS-I and 20fps in MESVS-II system.