A Scale Independent Selection Process for 3D Object Recognition in Cluttered Scenes

Abstract

During the last years a wide range of algorithms and devices have been made available to easily acquire range images. The increasing abundance of depth data boosts the need for reliable and unsupervised analysis techniques, spanning from part registration to automated segmentation. In this context, we focus on the recognition of known objects in cluttered and incomplete 3D scans. Locating and fitting a model to a scene are very important tasks in many scenarios such as industrial inspection, scene understanding, medical imaging and even gaming. For this reason, these problems have been addressed extensively in the literature. Several of the proposed methods adopt local descriptor-based approaches, while a number of hurdles still hinder the use of global techniques. In this paper we offer a different perspective on the topic: We adopt an evolutionary selection algorithm that seeks global agreement among surface points, while operating at a local level. The approach effectively extends the scope of local descriptors by actively selecting correspondences that satisfy global consistency constraints, allowing us to attack a more challenging scenario where model and scene have different, unknown scales. This leads to a novel and very effective pipeline for 3D object recognition, which is validated with an extensive set of experiments

Next-Best-View Method Based on Consecutive Evaluation of Topological Relations

“…The two test objects: On the left our Buddha statue, scanned with a high resolution 3D scanner and meshed. On the right, the Stanford Dragon meshed by (Rodolà et al, 2013 Mode) is averaged over 10 runs. The next four columns are the evaluation parameters from Section 3.4.…”

Learning to Rank 3D Features

“…Several examples of pose estimation using the proposed approach are shown in Figure 4. Note that Rodala et al [26] have also reported competitive results on Mian et al's dataset. However, in that study the recognition rate was evaluated using the feature matching accuracy rather than the pose estimation error, therefore it is not possible to directly compare with our method.…”

“…Various methods have been proposed. Many of them are based on selecting salient points in the 3D point cloud and using feature representations that can invariantly describe regions around the points [29,15,27,8,2,1,26]. These methods are known to produce successful results when the 3D object shape is rich and detailed, and the scene measurements are of high resolution and less noisy.…”

Efficient Shape Matching using Vector Extrapolation

“…The more recent techniques based on game theory [1,15] replace the unit norm constraint from (4) by the L 1 counterpart x 1 = 1, with x 0. The simplex constraint has the effect of promoting sparse, yet very stable solutions, a characteristic that makes the method particularly effective in tasks where strong selectivity is a major requirement [2,16]; nevertheless, this selectivity may come as a disadvantage in a variety of tasks, and the high locality of the obtained solutions does not allow, in practice, densification methods to be applied [23].…”

“…Arguably one of the most adopted formulations for shape matching takes form as a NP-hard quadratic assignment problem (QAP), where a quadratic term in the objective function encodes a measure of pairwise association among a set of putative matches. This formulation is common in attributed graph matching literature [1,7,9,23], but it also frequently arises in problems of inlier selection and object recognition [2,16], and in minimum distortion correspondence problems under the notion of Gromov-Hausdorff distance between metric spaces [12,14,15]. In this wide variety of scope and objectives, it is not uncommon for existing methods to provide specifically tailored optimization techniques for the particular problem they attempt to solve, ranging from dual decomposition [23] (6)).…”