Object detection of service robots is very important for their service. Most of services such as delivery, errand of users are related to objects. Conventional methods are based on the geometric models in static industrial environments, but they have limitations in uncertain and dynamic indoor environments, because interest object can be occluded or small in the image according to the robot's location or angle. For solving these uncertain situations, it is helpful to predict the probability of target object, because it can give important information for their next action. Our idea is to use observed objects as context information for predicting target one. For this, we adopt Bayesian networks and ontology together for modeling domain knowledge and reasoning objects in probabilistic frame. We verified the performance and process of our method through the experiments.