In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
will defend his proposal
Human Detection in the Wild
AbstractDespite the recent outstanding research progress in detecting humans in images, its applications of person re-identification, traffic safety, and autonomous driving pose a number of yet unsolved challenges. Existing human detection solutions focus on either detecting the human body or the face. Visible-body assisted approaches have been proposed to provide more reliable human information for the network to learn discriminative features. However, visible-body assisted approaches produce a large number of false positives, which result from a lack of adequate and discriminative contextual information of full-body and visible-body. Regarding face detection, the detectors could be categorized into two-stage, single-stage face detectors. The choice was driven by the computational requirements. In the two-stage face detectors, input pyramids or multi-scale feature maps are deployed to provide more reliable information of various scales of faces, especially faces of small size in an image, for the network to learn more discriminative facial features to detect faces in various scales. These actions increase the complexity of the network. In the single-stage face detectors, treating reliable information and noise equally could result in much noise in the fused features at different levels or leveraging stacked dilated convolutions to magnify receptive fields efficiently leads to the inconsistency of the local information. This proposal presents the following accomplishments: (i) Designed, developed, and evaluated a human detector for solving the occlusion problem, which achieves the state-of-the-art and promising results on the benchmarks. (ii) Designed, developed, and evaluated a two-stage face detector to reduce the complexity of the two-stage face detectors and achieve promising results of detecting multi-scale faces. (iii) Designed, developed, and evaluated a single-stage face detector to improve the performance of detecting multi-scale faces.
Date: Monday, December 09, 2019
Time: 10:00 AM - 12:00 PM
Place: HBS 317
Advisor: Dr. Ioannis A. Kakadiaris
Faculty, students, and the general public are invited.