Training Lightweight CNNs for Human-Nanodrone Proximity Interaction from Small Datasets using Background Randomization
This addresses the challenge of training lightweight CNNs for human-nanodrone proximity interaction with small datasets, though it is incremental as it builds on existing augmentation techniques.
The paper tackles the problem of visually estimating human pose from nano-drone images by proposing a data augmentation method using synthetic background substitution, which improves generalization to unseen environments as validated on data from two labs.
We consider the task of visually estimating the pose of a human from images acquired by a nearby nano-drone; in this context, we propose a data augmentation approach based on synthetic background substitution to learn a lightweight CNN model from a small real-world training set. Experimental results on data from two different labs proves that the approach improves generalization to unseen environments.