Altitude-Adaptive and Cost-Effective Object Recognition in an Integrated Smartphone and UAV System
Ignacio Martinez‐Alpiste, Gelayol Golcarenarenji, Qi Wang, José M. Alcaraz Calero · 2020
Human Search and Rescue (SAR) tasks are mission-critical and take place in the wild, and thus solutions require timely and accurate human detection on a highly portable platform. This paper proposes a novel lightweight and practical SAR system that meets those demanding requirements by running optimised machine learning in a smartphone, interoperable with Unmanned Aerial Vehicles (UAV) that provides live video feed. In particular, the proposed approach significantly extends a standard machine learning algorithm to achieve adaptive object recognition in response to changing altitudes to accelerate the speed of finding missing people and eliminate redundant computing. Our approach achieved 91.02% of accuracy and real-time speed on a smartphone that hosts the machine learning platform and the new algorithm. This proposed system is highly portable, cost-effective, fast with high accuracy suitable for UAV applications.