Smart Bioinspired Neuromorphic (Event-Based) Camera
Pixium Vision Bionic VRS features a unique proprietary neuromorphic bioinspired event-based camera, that mimics the human retina:Our eyes only see changes in our environment (contrast, intensity, luminance, etc.) and photoreceptors are activated independently.
Camera features and benefits:
• Neuromorphic - asynchronous, event-based: light is encoded into asynchronous impulses
• Breakthrough bioinspired camera: mimics normal vision in real time.
(-1,0,+1): Reduces energy consumption and processing bandwidth
• Output relates directly to signals observed in the corresponding levels of biological retinas:
Visual information submitted can then be directly understood by the visual cortex
In contrast to most other approaches in the field, Pixium is not deriving retina stimulation signals from gray-scale image frames acquired at arbitrary points in time (as done by conventional imagers) but from the output of a sensor that functionally emulates the human retina’s process of acquiring visual information from a scene.
The ATIS sensor contains an array of autonomous pixels each of which implement functional models of the two major human retina pathways, the Magno-cellular (or transient) pathway and the Parvo-cellular (or sustained) pathway. ATIS’ pixels are not driven by a shutter signal but, just like in the biological eye, respond to the visual input they receive from the scene. Pixel circuits encode transient (light change) information from the scene into the precise timing of spikes while sustained (light intensity) information is encoded using a simple spike rate coding scheme. The output of Pixium’s ATIS sensor is a continuous-time stream of spikes encoding transient and sustained visual information in a language the brain could directly interpret.
As a result of this biologically-inspired way of acquiring visual information, and in contrast to all conventional image sensors used today in vision restoration, ATIS is able to provide signals at the native temporal resolution of retinal ganglion cells (~1ms) and at a dynamic range exceeding the one of the human eye (>120dB). Also the impressive redundancy suppression performed by the human retina is reproduced by the sensor, making sure that only relevant and useful signals are sent to the brain.
The video above first illustrates the continuous-time acquisition of visual motion in a scene as performed by the ATIS sensor in contrast to a frame-based acquisition done by a conventional camera. Subsequently, the spike-encoded output from the two pathways is shown. The left-most screen displays the bipolar output of changes detected in the scene and acquired by the transient pathway pixel circuits. The middle screen shows the sustained pathway’s output encoding absolute luminance from the relevant parts of the scene in pixel-individual spike rates. The right-most screen shows the full scene content as acquired by ATIS. All visual information is encoded purely in the timing of spike pulses which constitute the basis of the stimulation signals sent to the retinal implant in Pixium’s vision restoration systems