A multi-processed salient point detection system for autonomous navigation

This paper proposes an unsupervised salient object detection system that able to extract potential exogenous regions of interests which may be used in robotic navigation system. Biologically inspired, this approach has novel implications for robotic vision that can reduce the complexity in global image processing by focusing only on representative basins of attentive attraction rather than every detail of the scene. This becomes especially critical when the robot navigates through large and sparse indoor spaces where global positioning is not available. Our approach combines three layers of processing, namely an initial scan by K-iterations fast learning artificial neural network color segmentation, followed by Gabor filtering to obtain edge information and finally the application of Scale Invariant Feature Transform (SIFT) for identification using prominent keypoints. A comparison is done with existing extraction methods and results are presented.