Insert beyond the traffic sign recognition: constructing an auto-pilot map for autonomous vehicles.

Traffic sign recognition (TSR) systems on the vehicles can collect posted speed limit sign information and have been in commercial usage since 2008. A daily-updated auto-pilot map can be constructed based on the massive amounts of TSR observations from multiple consumer vehicles; the data is then aggregated, filtered and processed, and the learned posted speed limit signs can be finally transferred to vehicles with high-coverage and real-time speed limit information. Compared with the direct sign detection by TSR systems, the auto-pilot map can complement the current detection errors, reduce the camera cost and provide a continuous speed limit information for autonomous vehicle applications. A pipeline of methods are specifically designed to deliver our research purpose by making full utilization of TSR observations and HERE map. Experimental results indicate that our proposed algorithms and methods can construct an auto-pilot map with an overall accuracy of 95.8%. It is also expected to update the speed limit information in a map at a faster pace than the traditional map since we are using sensors of customer vehicles instead of dedicated map construction vehicles. The utility of our proposed auto-pilot map opens a new perspective in autonomous driving.