How zenith angle of cross product affects NVP feature
Abstract
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Recent dramatic advance in visual recognition hardware has made some promising progress. For instance, the impact of the recent emerging low-cost depth sensors, i.e. the Kinect depth sensor, has extended far beyond the industry. This thesis targets the depth sensor based object detection task in images. We have been chosen to develop a robust feature extraction descriptor that is designed exclusively for depth images. Hence, we proposed a novel feature for object detection: Normal Vector Pattern (NVP). Generally, a superior feature should be discriminative enough to identify inter-class variations, whereas able to minimize the intra-class disturbance. The proposed NVP is successful due to its two components: the cross products which induce the discriminative power and the inner products which lead to the view-angle invariance. In this thesis, we further explore how different parameter settings may affect the performance of NVP feature.
Degree
M.S.
Thesis Department
Rights
Access is limited to the University of Missouri - Columbia.