Evaluation of PCL's Descriptors for 3D Object Recognition in Cluttered Scene

Abstract

3D point cloud recognition is a challenging problem that is growing fast, especially with the development of acquisition cameras such as the Kinect that allows the generation of a lot of data to approximate an object model. Based on the types of features used to represent an object from multi-views, 3D object recognition approaches can be classified into two broad categories which they use local or global features. The existing approaches often evaluate their performance with prior segmented 3D objects. In this paper, we propose a novel 3D object recognition pipeline based on different Point Cloud Library (PCL) descriptors as well as our proposed recognition threshold. Also, we use a prior object segmentation step after the real-world scene acquisition in order to extract all the objects which are present in it. Finally, the obtained experimental results should indicate the adequate algorithms to describe point clouds with respect to their time complexity and recognition rate.