Traffic Sign Recognition and fusion with navigation for reliable determination of current speed-limit

Being able to always know the current speed-limit is a highly interesting feature both for driver information system, and for advanced driving assistance such as a smart Adaptive Cruise Control (ACC) automatically adapting vehicle speed to current speed-limit. However, this kind of function can be really valuable and of practical use only if the produced speed-limit information is really reliable. In particular, it has to take into account roadwork temporary limits as well as variable speed limits enforced by LED signs, and also supplementary signs located below main signs and modifying their scope (class of vehicle, concerned lane, etc...). In order to attain the desired robustness level, the system has to rely not only on cartographic navigation information, but also on quite robust vision-based recognition of several kinds of traffic-signs, and also on proprioceptive information such as current speed, turn indicators, steering wheel angle, etc... In this paper, we present several of the required bricks for building such a reliable speed-limit support system: a performant recognition of speed-limit-signs working on various European roads (despite sign variability between country) and able to operate correctly also on LED signs and at night, good detection of end-of-speed-limit-signs, a method for effective detection and recognition of supplementary signs, and a general framework for intelligent fusion, using Belief Theory, of information deduced from vision and navigation. All these functions are implemented modularly, and have been evaluated both on real recorded videos, and by on-road real-time tests