Asynchronous pixels
Each pixel responds independently instead of waiting for a global frame readout.
Event-based vision / single-cell instruments
We are not slow. We are blind.
Argus is a C-mount event-vision module in bench development for single-cell instruments. Its architecture pairs per-pixel change sensing with on-module inference and a TTL gate output.
500 µs
Assumed transit through field
40 µs
Assumed time in decision region
<20 µs
Unvalidated target / sensor event to TTL edge
01 / Timing model
A frame can be longer than the event.
A frame camera samples the scene at fixed intervals. If a cell’s transit is shorter than that interval, a readout may contain no complete observation. Motion blur is a separate exposure-time effect.
Higher frame rates, reduced regions of interest, and shorter exposures can help, with trade-offs in field of view, illumination, or data rate. The model below isolates timing; its transit and gate values are assumptions, not Argus measurements.
Interactive timing model / 500 µs transit / 40 µs gate
One cell, two acquisition models.
Frame camera
Frame interval
—
Frame periods during transit
—
Modelled raw pixel rate / 8-bit mono
—
Sensor event → TTL edge
Unvalidated target: <20 µs
A retina reports local change.
Event sensors borrow that organising principle: pixels respond independently to brightness changes and emit timestamped events.
02 / Method
Report change instead of full frames.
Illustrative encoding comparison
The same moving signal, encoded two ways.
Frame model / 1280 × 720 / 8-bit mono
—
Every pixel is transmitted at every frame interval in this simplified model.
Event model / illustrative activity
—
This illustration marks pixels whose brightness changes beyond a simplified threshold.
Synthetic encoding model. Active-pixel percentage is not a bandwidth claim: event payload, noise, repeat events, and protocol overhead are excluded.
Sparse output
Event rate generally follows scene activity; static regions generate comparatively few events.
Timestamped events
Timebase granularity depends on the sensor and configuration; it does not establish end-to-end latency.
Log-intensity response
A logarithmic response helps retain changes across bright and dim regions.
03 / Instrument
Argus
Argus is being developed as a C-mount event-vision module for single-cell instruments. The intended closed loop connects an event sensor, on-module inference, and a TTL gate output.
- Modality
- Per-pixel brightness-change events
- Decision latency
- Unvalidated target <20 µs / event to TTL
- Mount
- C-mount / standard camera port
- Inference
- Designed for on-module execution
- Output
- Planned: event stream + TTL gate signal
- Status
- Subsystem bench validation / pilots recruiting
Argus system architecture
Conceptual architecture / not to scale
01
Change at pixel
Local log-intensity crosses threshold.
02
Timestamped event
Coordinates, time, and polarity.
03
On-module inference
Task-specific model evaluates the stream.
04
Gate output
Decision leaves as a TTL signal.
04 / Applications
Initial pilot focus: single-cell workflows.
We are recruiting pilot partners to test five application hypotheses in flow cytometry, slide scanning, live-cell imaging, microfluidics, and particle tracking.
Sorting & flow cytometry
Evaluate whether event streams can support classification and gate timing inside a short transit window.
Slide scanning & smear review
Test change-event acquisition during continuous stage motion; some assays may still require intensity reconstruction or frame fusion.
Live-cell dynamics
Evaluate long observations with sparse activity readout while retaining reference frames where absolute intensity matters.
Droplet & microfluidic assays
Measure per-droplet event-time distributions and compare them with the existing acquisition path.
Particle & motility tracking
Compare event-based trajectory reconstruction with interpolation between frame samples.
Tell us your timing constraint.
05 / Validation
Validation in progress.
Argus subsystems are in bench validation. The test plan covers event-to-TTL latency, task-specific classification, and event rate across sample densities. Results will be published with test conditions.
A labelled event-based cytology dataset is part of the planned work with pilot partners.
Decision-path latency
Distribution from the sensor event to the TTL output edge.
Classification performance
Task-specific comparison against an agreed frame-based baseline.
Event-rate envelope
Observed rate and maximum loss-free processing rate across sample density, illumination, and scene activity.
Technical brief in preparation: sensor selection, event-rate budget, and closed-loop timing.
Request the technical brief06 / Contact