A new Far Infrared (FIR) sensor from Israeli company AdaSky could help bring vehicles with Level 3, 4 and 5 autonomy features to market faster. The solution is the first to offer an FIR camera with built-in machine learning software for processing the results, in a solid-state form factor with no moving parts, at a cost that AdaSky says is "suited for mass market" use.
AdaSky's solution is called "Viper," and it works by passively collecting FIR signals sent out as heat by warm objects, including other cars and humans. Its range means that it can pick up signals from as many as a few hundred meters away, allowing it to detect pedestrians earlier than other sensors, and work in concert with different tech including LiDAR and traditional cameras to verify that pedestrians are in fact pedestrians ahead of time.
The high resolution thermal camera works both at a distance of a few hundred meters and up close, and it also helps fill in gaps in perception left by other solutions including vision cameras, radar and LiDAR – and even sensor fusion results from a combination of all three. Viper has advantages in changing lighting and weather conditions, for instance, where other sensors encounter problems, and it can add another dimension of signal from which to derive classifications of objects – offering more confidence that a pedestrian is a pedestrian, for instance, or that a person on a bike is a person on a bike.
AdaSky's Viper prototype has been heading out to tier-one auto industry suppliers and automakers quietly for a few months now, and the company hopes to move towards mass production and commercialization next.