The Visconti 4 image recognition processor is equipped with eight media processing engines, allowing it to execute eight applications simultaneously. It can detect and analyze camera-generated images and recognizes traffic lanes; nearby vehicles, both parked and moving; traffic signs and signals; the headlights of oncoming vehicles; plus the most vulnerable road users of all, bicyclists and pedestrians.

Visconti 4 has double the number of processing engines of its predecessor, the Visconti 2, used by DENSO since 2015. It also integrates a new image recognition algorithm—Enhanced CoHOG Accelerator—that delivers enhanced processing of luminance differences between objects and their backgrounds to better detect pedestrians at night and low light conditions. (CoHOG is Toshiba’s original Co-occurrence Histograms of Oriented Gradients technology.)

Toshiba and DENSO are also cooperating in AI, on the development of deep neural network-intellectual property (DNN-IP) for use in image recognition, and plan to bring this technology to future additions to the Visconti series.