Far Infrared cameras used initially for the driving of military vehicles are slowly coming into the area of commercial (luxury) cars while providing with the FIR imagery a useful assistance for driving at night or in adverse conditions (fog, smoke, ...). However this imagery needs a minimum driver effort as the image understanding is not so natural as the visible or near IR one. A developing field of FIR cameras is ADAS (Advanced Driver Assistance Systems) where FIR processed imagery fused with other sensors data (radar, ...) is providing a driver warning when dangerous situations are occurring.
The communication will concentrate on FIR processed imagery for object or obstacles detection on the road or near the road. FIR imagery highlighting hot spots is a powerful detection tool as it provides a good contrast on some of the most common elements of the road scenery (engines, wheels, gas exhaust pipes, pedestrians, 2 wheelers, animals,...). Moreover FIR algorithms are much more robust than visible ones as there is less variability in image contrast with time (day/night, shadows, ...). We based our detection algorithm on one side on the peculiar aspect of vehicles, pedestrians in FIR images and on the other side on the analysis of motion along time, that allows anticipation of future motion. We will show results obtained with FIR processed imagery within the PAROTO project, supported by the French Ministry of Research, that ended in spring 04.