Characterization of the High Speed Detection Limits of a Neuromorphic Vision Sensor
Abstract
Event-based sensors have demonstrated latencies on the scale of 110 microseconds, making the technology an energy-efficient, lightweight option for real-time, high-speed object detection and tracking. Research has been conducted to characterize the sensors abilities to detect high-speed objects in a controlled environment and establish performance upper limits and induce which on-chip channels and physical mechanisms act as bottlenecks. A configurable LED strip with user interface was designed and implemented to extract this detection cutoff in a lab-controlled environment. Since the event-based sensor responds to changes in intensity, switching on individual LEDs consecutively simulates the effect of an object moving across its field of view. A user interface was embedded into the LED strip to allow for customization of the simulated projectiles velocity. After calibrating the sensor, the detection limit was systematically extracted by increasing this velocity from 50 to 1000 m/s in rough increments. Using a custom signal-processing algorithm, the location of the projectiles tip was extracted. By tracking the number of average tip positions over the range of recorded shot velocities, linear extrapolation shows the absolute limit of the event-based sensor to be 1450 m/s. Faster speeds and more recordings can be used to verify this conclusion.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 19, 2020
- Accession Number
- AD1116559
Entities
People
- Aaron Bard
- Ben Linne
- Jonah Sengupta
Organizations
- United States Army Research Laboratory