IRT Smart Imager

Today, image sensing is largely deployed in a huge panel of applications. Smart sensing is the next step! Smart sensors are able to catch high level and comprehensive information from a scene, such as the presence of pedestrians on a road. They obviously integrate sensing and signal extraction but also processing to perform image and video analytics.
It is noticeable that smart sensing techniques benefited a lot from the silicon integration and 3D stacking. Thanks to the new frameless image sensor paradigm, the research field of image sensing is evolving very quickly. Contrarily to standard image sensors, which capture a scene at a fixed frame rate, event-based (frameless) image sensors only reacts to luminance changes at the pixel level. This means the output bitstream of the sensor is scene-dependent and anymore regular but intrinsically incorporates temporal and spatial coherence of the scene elements. Because of the very specific characteristics of this new sampling method, standard neural networks (NN) processing and up to date CNNs (Convolutional Neural Network) have to be translated to this new paradigm. It is expected that these new event based NNs are smaller and perform only the relevant processing, which will facilitate their silicon integration and lead to low power devices. Such a strategy will surely open the doors to new integrated sensing devices.