Handling Data Scarcity Through Data Augmentation in Training of Deep Neural Networks for 3D Data Processing

Due to the availability of cheap 3D sensors such as Kinect and LiDAR, use data in various domains manufacturing, healthcare, retail achieve operational safety, improved outcomes, enhanced customer experience has gained momentum recent years. In many these domains, object recognition is being performed using against difficulties posed by illumination, pose variation, scaling, etc present 2D data. this work, we propose three augmentation techniques for point cloud representation that sub-sampling. We then verify samples created through carry same information comparing Iterative Closest Point Registration Error within sub-samples, between sub-samples their parent sample, with different parents subject, finally, subjects. also augmented have characteristics features those original applying Central Limit Theorem.