In Partial Fulfillment of the Requirements for the Degree of Master of Science

Can Cao

will defend her thesis

An Interactive Pedestrian Re-identification Tool with Semantic Based Re-identification

Abstract

Person re-identification is an essential task of recognizing and matching people from non- overlapping cameras. A typical application of person re-identification is identifying a specific person in gallery pedestrian images from a camera with one or more given probe images of this person from another camera. It is a challenging and practical task that provides solutions for video-surveillance. In this work we present a person re-identification software which is called Interactive Pedestrian Re-identification GUI (IPRG), and a semantic-based labelling tool named Reid It (Reidit). We develop IPRG to address the person searching and matching problem. Given several semantic options, such as information of suspect as height, ethnicity, cloth color, etc. IPRG can recognize suspect and match him/her in the database. We also propose a light labelling tool Reidit for labelling pedestrian images with semantic feature as the pre- processing for pedestrian recognition. We present an experiment on IPRG with Viewpoint Invariant Pedestrian Recognition (VIPeR) dataset which contains 632 identities. Our experiment shows that our software is more efficient and accurate compared with traditional manual solutions. Moreover, IPRG can handle the situation of nonexistent query person in database, it will return top 10 candidates. Our software is compatible with different platforms and user-friendly with customizable database and semantic features.

Date: Tuesday, November 22, 2016
Time: 11:30 AM
Place: PGH 516
Advisor: Dr. Shishir Shah

Faculty, students, and the general public are invited.