Activity-Object Bayesian Networks for Detecting Occluded Objects in Uncertain Indoor Environment

Abstract. In the field of the service robots, object detection and scene understanding are very important. Conventional methods for object detection are performed with the geometric models, but they have limitations to be used in the uncertain and dynamic environments. This paper proposes a method to predict the probability of target object with Bayesian networks modeled based on activity-object relations. Experiments in indoor office environment show the usefulness of the proposed method for object detection, which produces about 86.5% of accuracy with environments.

• Type: Conference
• Year: 2005
• Where: KES
• Authors: Youn-Suk Song, Sung-Bae Cho, Il Hong Suh
• Keywords: KES 2005, Object Detection, Scene Understanding, Target Object