Distance Surface for Event-Based Optical Flow.

Area of Science:

• Computer Vision
• Robotics
• Sensor Technology

Background:

• Neuromorphic cameras, or event detection cameras, utilize dynamic vision sensors (DVS) to asynchronously report log-intensity changes (events).
• Traditional optical flow methods rely on pixel intensity, which is absent in event data from DVS cameras.

Purpose of the Study:

• To propose DistSurf-OF, a novel optical flow method specifically designed for neuromorphic cameras.
• To address the challenge of estimating motion using event data without direct pixel intensity values.

Main Methods:

• Introduced the concept of a "distance surface," computed via a distance transform from detected events, as a proxy for object texture.
• Integrated the distance surface as input into intensity-based optical flow algorithms to recover 2D pixel motion.
• Validated the method using real sensor experiments.

Main Results:

• DistSurf-OF accurately estimates the angle and speed of events from neuromorphic cameras.
• The proposed method demonstrates effective optical flow estimation in the absence of traditional intensity information.

Conclusions:

• The novel distance surface approach provides a viable solution for optical flow estimation with event-based sensors.
• DistSurf-OF represents a significant advancement for applications utilizing neuromorphic camera technology.