Active Object Recognition Based On Prior Feature Distribution Table

In this paper, an active object recognition (AOR) method based on a prior feature distribution table (PFDT) is proposed. According to the distribution information of predetermined features on each object in the model base, the prior feature distribution table is created. Taking it as input, a feature decision tree (EDT) is constructed for object recognition and viewpoint planning (VP). To determine the next best viewpoint, we transform it into an optimization problem that is solved with the tree dynamic programming algorithm. Experiments show that the proposed method can achieve the object recognition task with the minimum average number of viewpoints.