Multi-Camera Networks

CHAPTER 13 Online Learning of Person Detectors by Co-Training from Multiple Cameras

P.M. Roth, C. Leistner, H. Grabner, H. Bischof

Institute for Computer Graphics and Vision Graz University of Technology, Graz, Austria

Abstract

The detection of persons is an important task in automatic visual surveillance systems. In recent years, both, advanced representations and machine learning methods, have proved to yield high recognition performance while keeping false detection rates low. However, building general detectors applicable to a wide range of scenes leads to high model complexity and thus to unnecessarily complex detectors.

To alleviate this problem, we propose to train scene-specific object detectors by co-training from multiple cameras. ...