Vehicle Detection Onboard Small Unmanned Aircraft

Abstract

This book chapter presents a system and experiments for on-aircraft detection of ground vehicles. The focus is on the steps for creating an embedded real-time system that met operator desires despite the limitations of the computer vision methods, the communication network, and the small flight platform. Detailed experimentation was essential for planning and coordinating the hardware and software components to achieve real-time performance and the most benefit to the operators. A fast object detection method was repeatedly trained and evaluated until it achieved the desired speed and accuracy. The fault-tolerant client–server architecture delivered the most relevant information even despite severe netword degradation. The demonstrated content-aware filtering of imagery elevated the human analyst’s task from vehicle detection to detection verification, presumably a much less repetitive, fatiguing, and error-prone task. The system was successfully demonstrated as part of a Search And Rescue operation.