A Mobile Robot Application for 3D Object Classification via Point Cloud Data in Indoor Environment

3D object classification has been widely studied in robotic community in recent years. According to types of features that are utilized to describe an object, 3D object classification studies can be separated into two main categories (local and global).The local feature-based studies are robust against partial occlusion and clutter. However, they generally require more computational time and memory consumption when it is compared with global feature-based studies. On the other hand, global-feature based studies are appropriate for 3D shape classification because they observe the entire geometry of objects. In this study, a global feature-based approach, Viewpoint Feature Histogram (VFH), is used to classify table, chair and bookshelf objects. Support Vector Machine (SVM) is applied to classify features. To analyze the classification accuracy, we modelled the ESOGU Electrical Engineering Laboratory building in GAZEBO and a P3-AT mobile robot with RGB-D camera was used to construct Gazebo dataset that includes table, chair, and bookshelf objects. Similarly, a dataset for real-environment is constructed with the same objects in ESOGU Electrical Engineering Laboratory building. The test results show that the implemented method is able to classify each object with a classification rate above 83% and 60% in Gazebo dataset and realenvironment dataset, respectively.