CUVOS

Processing based on biology

Biological systems get reliable perception out of noisy, irregular receptors. It's the cells behind them that do the heavy lifting. Our neuromorphic processing is modelled on what those layers do.

Cuvos' technology takes individual sensor outputs and combines them across the frame and across time. Our systems don't use AI internally. None of the outputs contain information that was not provided by the sensor.

Across the sensor

Rebuilt using the elements beside it.

Across time

And what it read a moment before.

Processing as in nature

Neuromorphic engineering builds processing on the principles nervous systems use. Ours is modelled on the retina and the visual cortex, which turn noisy signals from millions of receptors into a stable picture.

Each layer of cells weighs a signal against its neighbours and against what it reported a moment ago. Noise that shows up in one place at one instant is discounted, and detail that holds across space and time is kept.

Cuvos applies the same principles to the output of a standard sensor. Working this way means processing is dynamic and operations can be applied as each sample arrives, helping meet tight timing constraints on low SWaP hardware.

Your sensorEach element reports only what it alone detected.Spatial correlationEach neighbour contributes.Temporal correlationPrevious samples from thatlocation contribute.ReconstructBoth readings combine into one value per element, from what thesensor captured.StructureEdges are kept where neighbours agree, then followed.Your processingDetection, tracking, recording or an operator.THE RETINACUVOS PROCESSINGPhotoreceptorsturn light into signalHorizontal & bipolar cellschanges in spaceAmacrine cellschanges in timeGanglion cellsspikes to the optic nervePrimary visual cortex (V1)oriented edges, contoursHigher visual cortexshapes, objects, meaning

Flexible implementation

Cuvos' technology can be added with minor disruption. Enhanced imaging will leverage the performance of your existing application.

  1. Your existing input

    Colour, thermal, hyperspectral. Raw or compressed.

  2. Cuvos

    Enhances the signal from its own history, in real time or offline.

  3. Your processing

    Detection, tracking, recording, or viewing. Better performance with your same methodology

Not just visible

Our approach can enhance the output of a variety of sensors. To date we have worked with thermal, hyperspectral, electro-optical, and RGB-NIR sensors. The same approach applies to other electro-optic systems and in principle to sonar.

Electro-optic

To date: thermal, hyperspectral, and NIR-RGB. In scope: full colour, low light, infrared, and ultraviolet.

Sonar

Future work: spatial and temporal sonar enhancement for passive, real-time detection and tracking.

Software, on CPU or GPU

The same algorithms run as an application, where throughput depends on the cores, memory, and bandwidth you can give it.

More detail available

We're always happy to engage in more in-depth discussions about our technology and where it might fit in your products.

Contact