|
|
Object |
 |
|
ETD: Multi-object tracking and associati... - Doctoral Dissertation (5 K) |
|
Creator |
Wu, Ziyan . |
Title |
Multi-object tracking and association with a camera network |
Degree |
PhD |
Department |
Dept. of Electrical, Computer, and Systems Engineering |
Is Part of |
Rensselaer Theses and Dissertations Online Collection |
Publisher |
Rensselaer Polytechnic Institute, Troy, NY |
Date |
2014-05 |
Language |
ENG |
Description |
May 2014 |
|
School of Engineering |
Abstract |
Video surveillance is a critical issue for defense and homeland security applications. There are three key steps of video surveillance: system calibration, multi-object tracking, and target behavior analysis. In this thesis we investigate several important and challenging computer vision problems and applications related to these three steps, in order to improve the performance of video surveillance. |
|
First, we introduce an airport security checkpoint surveillance system using a camera network. The system tracks the movement of each passenger |
|
and carry-on bag, continuously maintains the association between bags and passengers, and verifies that passengers leave the checkpoint with the correct bags. The real-time algorithms are validated on a full-scale simulation of a security checkpoint with several runs of volunteer groups, demonstrating high performance and efficiency in a challenging environment. |
|
Next, we propose a novel model for the internal and external calibration of Pan-Tilt-Zoom (PTZ) cameras that can cover the full range of pan, tilt and zoom. Based on this model, we propose a new self-calibration method for PTZ cameras. We then evaluate the drift in intrinsic and extrinsic parameters in a practical PTZ camera, and propose a dynamic correction method to maintain the correct calibration of a PTZ camera. Experimental results show that the proposed methods can keep an online PTZ camera calibrated, even over long surveillance sequences. |
|
Third, we investigate the detection of counter-flow in the exit lane of a US airport - that is, the abnormal motion of a person entering the secure area from the insecure area. We introduce a classifier to identify scene features in the image, which are used to mitigate cases in which foreground and background features are mixed in the same point trajectory, as well as to identify jitter frames that should not play a role in tracking. We demonstrate that our counter-flow detection algorithm is significantly improved by using the scene-feature-based classifier. |
|
Finally, we address two important problems related to human re-identification in camera networks. Since surveillance cameras are typically mounted high above the ground plane, we introduce a sub-image rectification method to eliminate the effects of perspective distortion. Moreover, we propose a model for human appearance as a function of pose what can be used in extracting roughly viewpoint invariant features for the images of the target and the candidates. We demonstrate superior performance on human re-identification applications in a real-world scenario. |
Contributor |
Radke, Richard J., 1974- |
|
Boyer, Kim L. |
|
Franklin, W. Randolph |
|
Cutler, Barbara M. |
Subject |
Computer and systems engineering |
Type |
Electronic thesis |
Rights |
This electronic version is a licensed copy owned by Rensselaer Polytechnic Institute, Troy, NY. Copyright of original work retained by author. |
Access Rights |
Restricted to current Rensselaer faculty, staff and students. Access inquiries may be directed to the Rensselaer Libraries. |
Record number |
000008074 |
Related collections |
|