Using the properties of an image for the purpose of camera fixation is a difficult task in robotics. In this research, the selection of pixels for use in a minimization process is discussed and compared against other possible strategies. The Slice Selection Process (SSP) which uses pixel displacement graphs to select the most suitable image region for camera fixation purposes has also been introduced. Choosing which slice to use is based on counting every graph peak and selecting the slice with the fewest peaks for camera fixation. We also present hierarchical search strategies that aim to perform as well as an exhaustive search with an enormous reduction in pixel processing. The performance of both the hierarchical search and SSP methods will be assessed in both laboratory and real world environments. Moreover, the difference in performance of various lenses with different angle views was tested to observe the effect of the large field of view on the ultimate outcome. The laboratory experiments were performed by hanging the platform over printed images to observe the system performance on various views. Nevertheless, to escape the constant environmental conditions in the laboratory environment (i.e. light/wind), real world experiments were performed to test the system performance under varying conditions. Some investigations were also carried out to study the influential factors (i.e. edges, regions, colours) on the ultimate results.
|Date of Award||05 Feb 2014|
|Supervisor||Mark Neal (Supervisor) & Frédéric Labrosse (Supervisor)|
Automatic Selection of Image Regions for Active Fixation
Hosseini, H. (Author). 05 Feb 2014
Student thesis: Doctoral Thesis › Doctor of Philosophy