| dc.description.abstract | The goal of this thesis was to implement a computer vision system on a low power platform, to see if that could be an alternative for a collision detection system. To achieve this, research into fundamentals in computer vision were performed, and both hardware and software implementation were carried out.
To create the computer vision system, a stereo rig were constructed using low cost Logitech webcameras, and connected to a Raspberry Pi 2 development board. The computer vision library OpenCV was used for interfacing with the stereo rig, and provide tools for image acquisition and stereo matching. The system was then put through a series of tests to evaluate the accuracy and refresh rate. For comparison reasons, an ultrasonic system were implemented with all the necessary hardware and software.
Based on the results obtained, a working implementation of a computer vision system was achieved. The system performed relatively good with respect to both the accuracy and the refresh rate, but will require some improvements depending on what type of applications the system are intended to be used in. The resulting computer vision system in this thesis does not provide any more functionality than the ultrasonic system in terms of collision detection, but it can be used as a foundation for future work to implement new features and more functionality.
The main contributions of this thesis is a computer vision subsystem that can be integrated for use into other applications directly. This system is also a foundation to be used for improvements and construction of a better system. | |
