Tracking People in Crowds by Feature Point Cluster Analysis Based on Spatial and Frequency Domain Cues

概要

論文の詳細を見る
The performance of detection and tracking systems greatly depends on the quality of its information sources. For tracking humans in crowded environments, this becomes especially important due to the additional challenges of heavy occlusions and the sheer number of target objects. Recent works based on feature point tracking and clustering have already begun to use motion cues in conjunction with spatial cues, and have been met with varying levels of success and improvement. We now propose the use of a third class of cues based on frequency domain analysis to enable a better understanding of human activity in crowded scenes. We present a system that takes advantage of multiple domains of information by using domain-specific methods such as motion trajectory and gait frequency analysis. The advantages of our method shall be shown as well through presentation of expenmental results.
一般社団法人情報処理学会の論文
2007-01-11