Analysis of the Adaptability of Key Point Detection Methods on 3D Point Cloud

Machine vision systems using 3D point clouds are becoming important for production management and quality control in the printing and packaging industry. Laying eyes on the effectiveness and utility of key point extraction, this paper has made comparison between the traditional hand-crafted algorithms and the deep learning algorithms, also including UAIP, one of our improved algorithms. Furthermore, the efficiency evaluation on key point extraction is from two aspects: quantitative and qualitative. They are in terms of semantic consistency, repeatability and stability. The algorithm adaptations in different contexts are also summarized in the paper. This experiment demonstrates the advantage of the learning-based detectors in key point detection with much more stable and repeatable results than the hand-crafted methods. The accuracy of our improved UAIP method has also been demonstrated better to a certain extent. This research clarifies the difference of key point extraction algorithms for 3D point clouds. With some practical insights on key point detection, it lays the foundation for the further research on point cloud shape.