Physical Reasoning for 3D Object Recognition Using Global Hypothesis Verification
In this paper, we propose a method to recognize the 6DoF pose of multiple objects simultaneously. One good solution to recognize them is applying a Hypothesis Verification (HV) algorithm. This type of algorithm evaluates consistency between an input scene
- PDF / 3,937,513 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 10 Downloads / 213 Views
Abstract. In this paper, we propose a method to recognize the 6DoF pose of multiple objects simultaneously. One good solution to recognize them is applying a Hypothesis Verification (HV) algorithm. This type of algorithm evaluates consistency between an input scene and scene hypotheses represented by combinations of object candidates generated from the model based matching. Its use achieves reliable recognition because it maximizes the fitting score between the input scene and the scene hypotheses instead of maximizing the fitting score of an object candidate. We have developed a more reliable HV algorithm that uses a novel cue, the naturalness of an object’s layout (its physical reasoning). This cue evaluates whether the object’s layout in a scene hypothesis can actually be achieved by using simple collision detection. Experimental results show that using the physical reasoning have improved recognition reliability. Keywords: Hypothesis Verification detection · Point cloud
1
· 3D object recognition · Collision
Introduction
3D object recognition and 6DoF pose estimation from depth data is one of the fundamental techniques for scene understanding, bin-picking for industrial robots, and semantic grasping for partner robots. A Model-based Matching (MM) method is generally used to recognize 6DoF of target objects. The MM detects 6DoF pose parameters having high fitting score exceeding a predefined threshold. However, this method sometimes detects false positives on surfaces having a high fitting score. The reasons are the occurrence of a pseudo surface shared by multiple objects and the presence of objects having parts similar to those of the target object. This is the essential problem of the MM approaches because they detect objects on basis of local consistency such as the surface of object model and the partial regions of the input scene data. One good solution to solve this problem is applying the Hypothesis Verification (HV) algorithm shown in Fig. 1. HV is an algorithm for scene understanding that uses the MM algorithm as a module for generating object candidates. c Springer International Publishing Switzerland 2016 G. Hua and H. J´ egou (Eds.): ECCV 2016 Workshops, Part III, LNCS 9915, pp. 595–605, 2016. DOI: 10.1007/978-3-319-49409-8 51
596
S. Akizuki and M. Hashimoto
Fig. 1. The overview of the HV algorithm for multiple object recognition.
This algorithm simultaneously recognizes the 6DoF poses of multiple objects by matching an input scene and the scene hypotheses represented by combinations of object candidates generated by the MM with low thresholds. Using this algorithm enables reliable recognition. This is because it maximizes the consistency calculated from global information, such as consistency between the input scene and the scene hypothesis, instead of maximizing the local consistency. Using the HV algorithm helps to achieve reliable recognition; however, it sometimes detects spatially overlapping objects. This is because it evaluates 2.5D consistency, the similarity of scene point cloud and rend
Data Loading...
