UVM Theses and Dissertations
Format:
Print
Author:
Van Dam, Clark
Dept./Program:
Electrical Engineering
Year:
2014
Degree:
Ph. D.
Abstract:
Visual tracking and localization are seen to be dual problems in 2-Diandto-D space; both are critical tasks for application to planetary rover exploration. Localization may be applied to improve tracking and vice-versa.
Here, 3-D measurements are utilized to improve 2-D feature-based visual tracking algorithms via guiding template update. 2-D subspace projections of tracking results within a known 3-D environment are used to effect monocular localization and automatically detect occlusion duration. Drift is reduced in 2D tracking by making use of 3D occlusion detectors to automatiocally select template update paths which avoid update during occluding events. Localization is improved using feature correspondences from simulataneous 2D target locks in combination with triangulation.
Here, 3-D measurements are utilized to improve 2-D feature-based visual tracking algorithms via guiding template update. 2-D subspace projections of tracking results within a known 3-D environment are used to effect monocular localization and automatically detect occlusion duration. Drift is reduced in 2D tracking by making use of 3D occlusion detectors to automatiocally select template update paths which avoid update during occluding events. Localization is improved using feature correspondences from simulataneous 2D target locks in combination with triangulation.