3D Modelbased Detection of People in Monocular Video Sequences taken in Trainstations

摘要

This article focuses on a new approach for three-dimensional people detection in monocular image sequences taken from stationary cameras with fixed focal lengths. The main contribution of the new algorithm is the combination of the following four parts. An abstract articulated geometrical person model, a scene model including a platform and a camera model, a hypothesis generator for people positions based on the CONDENSATION algorithm, and a fast clustering approach for the people detection. This approach allows the robust detection of several persons on the platform in realtime. This approach demonstrates that even with monocular image processing algorithms it is possible to detect people in realtime and to obtain information on their three-dimensional locations.