Objects Relationship Modeling for Improving Object Detection Using Bayesian Network Integration

Object detection is very important to service robots. Many tasks for service such as delivery, cleaning, and health-care for elderly people are strongly related to objects. Conventional approaches for object detection are mainly based on the geometric models, because they have been applied to static environments. In indoor environments having uncertainty, they have limitation in some situations where interesting objects are occluded by other ones or small in the scene. Context information can be helpful to overcome these uncertain situations. In this paper, we adopt objects as context information to allow for service robots to predict the probability of interesting objects through observed ones. For this, an object relationship model based on Bayesian network (BN) and integration method are proposed. Experimental results confirm that the proposed method predicts the objects very well.