Object co-occurrence graph for object search in 3D environment

We propose a method of using object co-occurrence graph for mobile service robots to search for small-scale objects in 3D environments without pre-defined map. Object co-occurrence graph describes co-occurrence relations between objects in scenes. The presence of other large objects in the environment, which are easier to be found, provides cues to search for a small object related to the large object. If the target object is often near to a large object, robot firstly finds the large object, and then searches around the large object for the target object. Thus, using object co-occurrence graph makes object searching task easier and more efficient. The object co-occurrence graph is automatically constructed by many tagged images from the WWW. Our experiment shows a robot searching for a target object using object co-occurrence graph.