Recognizing objects in 3D point clouds with multi-scale local features.

Min Lu,Yanxin Ma,Yinjie Lei,Yulan Guo,Jun Zhang

doi:10.3390/s141224156

Abstract

Recognizing 3D objects from point clouds in the presence of significant clutter and occlusion is a highly challenging task. In this paper, we present a coarse-to-fine 3D object recognition algorithm. During the phase of offline training, each model is represented with a set of multi-scale local surface features. During the phase of online recognition, a set of keypoints are first detected from each scene. The local surfaces around these keypoints are further encoded with multi-scale feature descriptors. These scene features are then matched against all model features to generate recognition hypotheses, which include model hypotheses and pose hypotheses. Finally, these hypotheses are verified to produce recognition results. The proposed algorithm was tested on two standard datasets, with rigorous comparisons to the state-of-the-art algorithms. Experimental results show that our algorithm was fully automatic and highly effective. It was also very robust to occlusion and clutter. It achieved the best recognition performance on all of these datasets, showing its superiority compared to existing algorithms.

Highlights

Object recognition is an active research topic in the area of computer vision [1,2]
It can be inferred that the proposed multi-scale rotational projection statistics (RoPS) feature-based algorithm further improves the performance of 3D object recognition compared to the fixed-scale RoPS
In order to have a fair comparison with the results reported by exponential map (EM), variable-dimensional local shape descriptors (VD-LSD), 3DSC, spin image- and fixed-scale RoPS-based algorithms, we tested our coarse-to-fine 3D object recognition algorithm on the same dataset as [4,19,32]

Summary

Introduction

Object recognition is an active research topic in the area of computer vision [1,2]. It has a number of applications, including robotics, forensics, surveillance and remote sensing [3,4,5]. Global feature-based algorithms describe the whole surface of an object by a single descriptor They require the scene point cloud to be pre-processed by a suitable 3D segmentation algorithm for the purpose of extracting individual object instances in the presence of clutter and/or occlusions [13]. They are frequently investigated in the area of shape classification and model retrieval [14]. Local feature-based algorithms have attracted more interests due to their robustness to clutter and occlusion [11,15,16] They first identify a number of keypoints in a scene and extract a feature descriptor for each keypoint. These feature descriptors of the scene are matched against these feature descriptors of 3D models to get the recognition results

Objectives

Results

Conclusion