People tracking in sparse crowd using multiple cameras
Jan Zimmermann · Digital Repository (National Repository of Grey Literature) · 2010
In this thesis, we introduce a new algorithm for tracking of multiple people in image sequences from surveillance cameras. The algorithm handle with trustworthiness in similarity measure of human gures in images. All decisions are made as trustworthily as possible. The algorithm nds out the number of people in the scene autonomously. It deals with potential temporary occlusions of people in image. The tracking algorithm does not suffer from the problem of human model initialization as some other methods do. The implementation of our algorithm is a part of this thesis. This implementation has veri ed the characteristics of the algorithm. We have used two different methods for the image segmentation. Both method prove successful in different conditions. We have used a color information to compare the visual similarity. Suitable future extensions are suggested at the end of this thesis.