Recognition of Point Sets Objects in Realistic Scenes

Ruizhen Gao,Jingjun Zhang,Xiaohui Li

doi:10.1155/2020/4925821

Abstract

With the emergence of new intelligent sensing technologies such as 3D scanners and stereo vision, high-quality point clouds have become very convenient and lower cost. The research of 3D object recognition based on point clouds has also received widespread attention. Point clouds are an important type of geometric data structure. Because of its irregular format, many researchers convert this data into regular three-dimensional voxel grids or image collections. However, this can lead to unnecessary bulk of data and cause problems. In this paper, we consider the problem of recognizing objects in realistic senses. We first use Euclidean distance clustering method to segment objects in realistic scenes. Then we use a deep learning network structure to directly extract features of the point cloud data to recognize the objects. Theoretically, this network structure shows strong performance. In experiment, there is an accuracy rate of 98.8% on the training set, and the accuracy rate in the experimental test set can reach 89.7%. The experimental results show that the network structure in this paper can accurately identify and classify point cloud objects in realistic scenes and maintain a certain accuracy when the number of point clouds is small, which is very robust.

Highlights

Point cloud is a collection of points
Using the deep learning network structure proposed in this paper to directly recognition point cloud objects can greatly reduce the amount of data calculation
Point clouds do not introduce quantization artifacts, which can better maintain the natural invariance of data. e experimental results show that the network structure in this paper can accurately identify and classify point cloud objects in realistic scenes and maintain a certain accuracy when the number of point clouds is small, which is very robust

Summary

Introduction

Point cloud is a collection of points. It contains rich information, which can be three-dimensional coordinates X, Y, Z, color, intensity value, time, and so on. We use deep learning network structures to perform feature extraction and recognition for each point cloud object in a realistic scene. Using the deep learning network structure proposed in this paper to directly recognition point cloud objects can greatly reduce the amount of data calculation (compared to mainstream methods such as converting point clouds to regular depth maps, multiviews, or voxel grids). One is to convert the point cloud data into a multiview, polygon network, or voxel grid and use deep learning networks for feature extraction and recognition (as described above). CNN can extract high-level semantic information from the original data through a series of operations such as convolution and pooling and generate a valid feature It aims to improve the classification accuracy of large-scale multicategory complex 3D models. On the Generation of Point Cloud Data Sets: Step One in the Knowledge Discovery Process [20], Andreas Holzinger et al describe the case for natural point clouds and provide some fundamentals of medical images, dermoscopy, confocal laser scanning microscopy, and totalbody photography; they describe the use of graph theoretic concepts for image analysis, give some medical background on skin cancer, and concentrate on the challenges when dealing with lesion images and the discussion of related algorithms. e point cloud data is extracted from different weakly structured sources and topologically analyzed to produce feasible results. e quality of these results depends on the quality of the algorithm itself, and to a large extent on the quality of the input maps they receive, so point clouds are a necessary preprocessing step and affect the quality of the experimental results

Problem Statement

Euclidean Distance Clustering Segmentation and Data Preprocessing

Method Time

Deep Learning on Point Sets

Experiment