Defining Performance Metrics for Closed Loop Event Based Imaging Systems
Supervisors
Dr Nimrod Kruger & A/Prof Gregory Cohen
Description
The current focus on event-based sensing in academia and industry revolves around creating advanced sensors, characterizing them, and developing data analysis pipelines. There are a plethora of exciting demonstrations highlighting this technology’s capabilities, but benchmarks and comparative performance metrics against traditional imaging modalities are scarce. Simple questions, such as “At what distance can an event camera detect a drone?” cannot be answered without an extensive design, prototyping, and data collection cycle. A design process for a conventional camera includes performance metrics and criterions assessed before any hardware is purchased or lines of code are written. Without the equivalent for event-based sensing, their adoption for high-end tasks is delayed.
This project aims to define the metrics and criterions for event-based imaging systems performance. Through several design-to-prototype cycles of unique designs, we will establish what constitutes a “good” event sensor and event-based imaging system, setting new standards for manufacturers of sensors, cameras, and systems.
Outcomes
- Survey current state-of-the-art in vision performance metrics for both conventional and neuromorphic sensors.
- Define specialized metrics for event-based signals fitting closed loop processing (line of sight motion) and dynamic scenes.
- Evaluate these metrics, through various analytical tools and literature, to assess performance – and establishing design criterions.
- Construct a performance model: a designer tool used for evaluating performance of imaging system at the drawing board stage.
- Initiate a full design process for a unique, high-end, event-based imaging systems through several prototype stages.
