Bio-Inspired Visual Navigation: From Landmarks via Bearing to Controls

In a low-texture indoor environment, landmark-based navigation is difficult due to lack of distinctive micro-features such as SIFT and SURF. Intelligent Automation, Inc. (IAI) and GRASP Laboratory at the University of Pennsylvania propose a Bio-Inspired Visual Navigation System with the following three innovations: First, we propose a segmentation-based method to extract landmarks as closed contours in the image which are globally salient and can be detected even in the low-texture environment. We then extract view-invariant shape and topological features that can be used to recognize these landmarks from different viewpoints. Second, we transform the locations of these landmarks into visual guides for bearing-only navigation. Third, we propose an intuitive OCU user interface for the operator to specify the waypoints and destination for the UGV. Using this new type of landmarks, UGVs will be able to navigate its surrounding to the required destination with minimal inputs from the operator. The proposed solution can be integrated into Armyfs existing UGVs and would allow the UGVs to operate in non-line-of-sight teleoperation with degraded communications.