EVPropNet: Detecting Drones By Finding Propellers For Mid-Air Landing And Following

Event: CVPR 2025 · Duration: 2 min · ▶ Watch on YouTube

Abstract

This video presents EVPropNet, a novel method that utilizes event camera data to detect drone propellers for advanced robotic applications. Event cameras, unlike standard cameras, capture asynchronous intensity changes, making them highly effective at handling fast movements and avoiding motion blur, which is crucial for detecting rapidly spinning drone propellers. EVPropNet processes this event data to accurately identify propellers. The paper demonstrates two key applications: tracking and following an unmarked drone, and performing mid-air landing on a near-hovering drone, showcasing the robustness and precision of the event-based detection system.

Speakers

  • Nitin J. Sanket — University of Maryland (UMD) PRG
  • Chahat Deep Singh — University of Maryland (UMD) PRG
  • Chethan M. Parameshwara — University of Maryland (UMD) PRG
  • Cornelia Fermüller — University of Maryland (UMD) PRG
  • Guido C.H.E. de Croon — TU Delft MAVLab
  • Yiannis Aloimonos — University of Maryland (UMD) PRG

Talks (1)

  • 00:00:00 — Nitin J. Sanket: EVPropNet: Detecting Drones By Finding Propellers For Mid-Air Landing And Following
    • This paper introduces EVPropNet, an event-based vision system for detecting drone propellers to enable robust mid-air landing and tracking of unmarked drones.

Key Takeaways

  • Event cameras offer significant advantages over standard cameras for detecting fast-moving objects like drone propellers due to their high temporal resolution and immunity to motion blur.
  • EVPropNet is a specialized neural network designed to process event data for robust and precise propeller detection on drones.
  • The propeller detection capability of EVPropNet enables advanced drone applications such as tracking and following unmarked drones, and performing complex mid-air landings.
  • The research highlights the potential of event-based vision for real-time, high-speed robotic perception tasks.

Methods / Models / Datasets Mentioned

  • EVPropNet
  • DVS (Dynamic Vision Sensor)

Topics

Drone detection · Propeller detection · Event cameras · Event-based vision · Mid-air landing · Drone tracking · Drone following · Asynchronous sensing · Robotics

Notes

Open for commentary — connections to other work, critiques, follow-up reading.