Vision-based indoor scene cognition using a spatial probabilistic modeling method

Abstract

This work describes a vision-based approach to recognize scene in the indoor environment. The proposed method represents each scene captured by a Pan-Tilt-Zoom (PTZ) camera with a blob model using spatial probabilistic modeling. Although the details of the scene covered by the camera are lost, this model is efficient in memorizing the scene characteristics and is robust against image distortions. Furthermore, multi-view recognition is studied to increase the precision of scene cognition via a partial knowledge of the scene. The images captured in the same location with different view angles are collected to extract the scene characteristics in order to decrease the memory storage size for each location. The effectiveness of the method is demonstrated by experiments in an unstructured indoor environment.