Training Lightweight CNNs for Human-Nanodrone Proximity Interaction from Small Datasets using Background Randomization

TL;DR

This paper introduces a background randomization data augmentation method to train lightweight CNNs for human pose estimation from nano-drone images, effectively improving generalization from small datasets.

Contribution

It presents a novel background substitution technique for data augmentation, enabling effective training of lightweight CNNs with limited real-world data.

Findings

Improved generalization to unseen environments

Effective training with small datasets

Validated on data from two labs

Abstract

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.