Image Recognition Algorithm of Front Collision Warning System using Monocular Camera for CA NCAP

摘要

Collision Avoidance NCAP (CA NCAP) was issued towards reducing traffic accidents in theUSA. Cost reduction is essential for spreading these systems in the market and making thisprogram effective. A multi-function front camera system appears promising, which canrealize two functions of FCW and LDW in CA NCAP by one sensor. However, obstaclerecognition applying image processing is still a difficult challenge due to the severeillumination condition in general outdoor environment. Furthermore, large calculation amountis necessary for the recognition processing, which makes hardware expensive. We havedeveloped efficient algorithms based on stochastic concept in order to resolve thesechallenges. Optical flow combined with pattern identification based on statistical probabilityis adopted to improve the recognition robustness. And, structural vehicle pattern classifierwas designed to make adjustments easy in case of recognition errors.