Remanent Imaging
Conventional video samples motion in brief snapshots. Remanent Imaging integrates it: long exposures at a low frame rate record each moving animal as a continuous light trail, so motion can be read directly from the images.
Motion recorded as light trails
Near-infrared light, invisible to the fish, crosses the tank from the side. The camera sees only the light scattered by the animals, so they appear as bright trails on a black background, day and night, without disturbing them.
This works even for small, transparent animals such as zebrafish larvae. Successive images join seamlessly, and stacking them builds a complete record of every movement: in the video, 20 seconds of motion by 12 larvae build up into trajectories that can be followed one by one.
The sensor is almost always collecting light
In CanaryTank, Remanent Imaging records at about 3.75 images per second with exposures of about 0.266 s. The interval between images is 1/3.75 s ≈ 266.7 ms, so the sensor collects light about 99.7 % of the time; the gap between exposures is under a millisecond.
As a result, anything that makes a fish move, such as a banging door, a flash of light or a vibration, is recorded, however brief.
A conventional camera behaves differently. At 30 images per second with a 1 ms shutter, for example, the sensor collects light only 3 % of the time, and anything that happens in the remaining 97 % can be missed.
Remanent Imaging also needs far fewer images: motion is measured from a single image rather than by comparing two or more.
| Remanent Imaging | Conventional video | |
|---|---|---|
| Capture principle | Integrative: long exposures, contiguous in time | Sampling: brief exposures separated by gaps |
| Brief events | Recorded: a fish's reaction appears as a change in its trail | Can fall between frames and be missed |
| Motion measured from | A single image (the trail is the motion) | At least two consecutive images |
| Images per day | 324,000 at 3.75 images per second | 2,592,000 at 30 images per second (8× more) |
| Real-time analysis | Each image analyzed on its own, so analysis keeps up in real time | Heavier, sequential processing |
| Long-term storage | Years of continuous recording are practical | Storage cost limits recording duration |
Mechanistic and inspectable
Trajectories and measurements are extracted from the images by a transparent geometric method, not by a trained black-box model. The same images always give the same results, and anyone can check a measurement by looking at the image it came from, because the motion is visible in it.
This matters for scientific rigor and reproducibility: results can be audited, explained and reproduced. It also makes the data a reliable, human-verifiable ground truth for training AI models.
Real-time processing of the image stream is made possible by our patented Remanent Imaging Control. Patents →