Are Current 3D Descriptors Ready for Real-time Object Recognition?

3D object recognition based on local features is a rapidly growing research field. For real-time object recognition, training free techniques are always preferred as they are free from heavy statistical learning. This paper presents a critical analysis of 3D texture-less object recognition techniques that are free from any training. Top-rated training free recognition techniques such as Spin, SHOT, RoPS, FPFH and RSD are evaluated on different types of datasets such as synthetically constructed and dataset acquired by an RGBD camera. We also present a dataset of five objects to analyze the performance of considered techniques. Based on our experimentation, we discuss the applicability of recognition techniques in real-time and also present discussion on future research directions.