Brookshaw, Iain (2011) Real time implementation of obstacle avoidance for an autonomous mobile robot using monocular computer vision. [USQ Project]
![[img]](https://sear.unisq.edu.au/style/images/fileicons/application_pdf.png)
|
PDF
Brookshaw_2011.pdf Download (2MB) |
Abstract
For autonomous robotic motion, it is essential for the mobile machine to be able to judge its position relative to potential obstacles. This implies the ability to identify potential obstacles, study their approach and take appropriate action when necessary to avoid collision. In this age of cheap and available digital cameras and powerful computers, it is desirable to achieve this with a single digital camera as the sensor.
The digital camera has the advantages that it is cheap, easily installed, well understood, passive and physically small. When coupled with a small, powerful on-board computer it has the potential to create a effective obstacle avoidance system.
This investigation was attempt to create just such a system. A series of methods dealing with identifying, tracking and avoiding obstacles were be investigated,
and the results of their implementation discussed. Alternative methods were be analysed and weighed. It was the initial intention to produce a program that could identify parts of the image as 'obstacles', determine the approach of these obstacles and use this information to direct a small mobile, autonomous machine.
To achieve this a region based segmentation approach was used to identify the boundaries and extents of obstacles, while several Looming based methods were employed to judge object range and approach, specifically Looming through blur
and Looming through area. Also considered were the ways and means of creating frame to frame correlation of objects based on region geometry. Unfortunately, the final result was too inconsistent to enable true avoidance to be implemented.
The inconsistencies in the results were largely due to small errors in each section compounding as the program evolved. However, a wide range of tests on each of the component parts of the system illustrated that the concepts and methods selected are viable individually. While there were problems with consistency when the program is run, it was clear from the results that the individual components
could be made to function if additional time was spent correcting errors.
The end conclusion was that, although the methods discussed were clearly viable, they require further experimentation and development before they can be fully implemented.
|
Statistics for this ePrint Item |
Actions (login required)
 |
Archive Repository Staff Only |
