Abstract:
Inspecting power networks using autonomous unmanned aerial vehicles (UAVs) has gained significant attention due to rapid advances in embedded devices, such as Jetson, and UAV technology. UAVs equipped with high-end onboard processing units and camera payloads are dispatched across the network to acquire high-quality data quickly and safely. This is particularly challenging when the location of infrastructure components, like poles, is unknown. In this work, we capitalize on breakthroughs on Jetson devices to develop a vision-based AI toolkit, which can process sensory input from the UAV’s camera payload in real time, and detect poles whose location is unknown. Detection output is integrated with the flight controller for aligning the UAV directly above the pole, marking its correct location. The proposed approach has successfully inspected about 3.5 kilometers of a medium-voltage network in an unseen region.
Nvidia Publication – Poster – Video,