NICOP - Project: Online Multi-Person Tracking with Two-Stage Data Association and Online Appearance Model Learning
Abstract
NICOP Project: Online Multi-Person Tracking with Two-Stage Data Association and Online Appearance Model Learning:The goal of this research part is to develop a novel online multi-person tracking algorithm based on sequential track-by-detection framework which can be applied to online/real-time applications. The novel online appearance model learning addresses the following two issues: (i) online learning to update appearance models based on online tracking results and (ii) online training sample collection for discriminating appearances of people and backgrounds in real time. Most previous tracking methods with appearance model learning focus on only one of the two issues. The problem of automatic multi-person tracking in computer vision has received much interest from researchers in recent years. The multi-person tracking concerns estimating the locations and sizes of multiple persons and conserve their identities (IDs). It is important for many applications including as diverse sectors as security, defense, robotics as well as communications. In complex scenes, this problem is quite challenging because there are frequent and prolonged occlusions by people interactions or obstacles and abrupt motion and appearance change of people. A benefit to developing an appearance model is in re-identification of subjects who reappear in the scene. This is especially important for recognizing potentially suspicious behavior, and can be extended to include identification from one camera to another for continuous tracking in a secure, multi-camera surveillance system. This work should improve the usability of surveillance data for finding the same subject like the Boston Marathon Bomber Tsarnaev much more quickly and easily. The use of video with acoustic combines cheap sensors that can be easily deployed but with a realistic fusion of technology to improve the overall system.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 23, 2016
- Source ID
- N629091612185
Entities
People
- Hanseok Ko
Organizations
- Korea University
- Office of Naval Research
- United States Navy