Feedback control of event cameras

Event: T. Delbruck, 2021 (presentation date 6/19/2021) · Duration: 3 min · ▶ Watch on YouTube

Abstract

This presentation details a novel feedback control mechanism for event cameras, specifically Dynamic Vision Sensors (DVS). The system dynamically modifies internal pixel parameters such as event threshold, bandwidth, and refractory period by monitoring the statistics of the outgoing event stream, including event rate and signal-to-noise ratio. This autonomous regulation prevents event stream saturation and optimizes data output for varying scene activity. A live demonstration using a DAVIS346Blue camera showcases how the refractory period is adjusted to maintain a target event rate, highlighting the utility of this approach for applications in IoT and mobile robotics where continuous human control is impractical.

Speakers

  • Tobi Delbruck — Inst. of Neuroinformatics, UZH-ETH Zurich, Switzerland
  • Rui Graca — Inst. of Neuroinformatics, UZH-ETH Zurich, Switzerland
  • Marcin Paluch — Inst. of Neuroinformatics, UZH-ETH Zurich, Switzerland

Talks (1)

  • 00:00:00 — Tobi Delbruck: Feedback control of event cameras
    • This talk introduces a feedback control system for event cameras that dynamically adjusts pixel parameters like bandwidth, threshold, and refractory period to regulate the event stream output based on real-time statistics.

Key Takeaways

  • Event cameras can be equipped with internal feedback control to dynamically adjust pixel parameters like bandwidth, threshold, and refractory period.
  • This control mechanism uses real-time statistics of the event stream (e.g., event rate, signal-to-noise) to regulate the camera’s output.
  • Dynamic adjustment prevents event stream saturation and optimizes data flow, especially in high-activity scenarios.
  • The demonstration shows effective control of the refractory period to limit the event rate, maintaining a stable output.
  • Such autonomous feedback control is highly beneficial for applications in IoT and mobile robotics where human intervention in the camera’s operation is not feasible.

Methods / Models / Datasets Mentioned

  • DVS camera
  • DAVIS346Blue mini camera
  • JAEAR
  • EventRateEstimator
  • SpatioTemporalCorrelationFilter

Topics

Event cameras · DVS · Feedback control · Adaptive biasing · Event rate regulation · Refractory period · Bandwidth control · Threshold control · Neuromorphic vision · Real-time processing

Notes

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