Attention and Sensor Planning in Autonomous Robotic Visual Search

This thesis is concerned with the incorporation of saliency in visual search and the development of sensor planning strategies for visual search. The saliency model is a mixture of two schemes that extracts visual clues regarding the structure of the environment and object specific features. The sensor planning methods, namely Greedy Search with Constraint (GSC), Extended Greedy Search (EGS) and Dynamic Look Ahead Search (DLAS) are approximations to the optimal solution for the problem of object search, as extensions to the work of Yiming Ye. Experiments were conducted to evaluate the proposed methods. They show that by using saliency in search a performance improvement up to 75% is attainable in terms of number of actions taken to complete the search. As for the planning strategies, the GSC algorithm achieved the highest detection rate and the best efficiency in terms of cost it incurs to explore every percentage of an environment.