Vehicle Recognition and TTC Estimation at Night based on Spotlight Pairing

摘要

Forward Collision Warning (FCW) together with Lane Departure Warning (LDW) and Electronic Stability Control (ESC) can highly contribute to a safer traffic environment. Therefore those three systems will be supported by the US NCAP (New Car Assessment Program) until the end of 2009 and rewarded in the new active safety rating. LDW and ESC are already on the market whereas development on the first vision-based FCW systems is still ongoing and enforced. This paper will give a short overview on common approaches in vehicle recognition and focus on nighttime conditions. Based on existing head- and taillight detection algorithms for Adaptive Headlamp Control (AHC) a novel vehicle detection algorithm is presented. The main challenge hereby is the accurate pairing of detected vehicle spots. The developed system also contains rough distance estimation and a robust Time-To-Collision (TTC) calculation. Sequences in various traffic conditions with hand-labeled vehicles serve as an evaluation set for the vehicle detection performance. Ground truth for TTC is available by radar sensor data. The system proves to be robust in various near-collision situations.