Tensor voting based road lane recognition algorithm with geometric constraints for lane departure warning system

摘要

In view of the restricted performance in terms of accuracy, anti-disturbance, and robustness of the current road lane recognition algorithm for Lane Departure Warning System, tensor voting–based road lane recognition algorithm with road lane geometric constraints is presented in this paper. It extracts road lane information from random images using voting decayed function optimized from geometric constraints of road lanes. The entire framework of the algorithm consists of pre-processing, vote-processing, and post-processing. First, white points which represent the potential road lanes are selected and remaining points are discarded for reducing noise and computation burden in binary image. Next, the most likely connection between voters and receivers can be characterized as decayed function. Road lane geometric constraints determined by calibration results of the camera are utilized to optimize decayed function in tensor voting in order to emphasize road lane points in vote-processing. At last, these points are clustered and fitted into road lanes in post processing. The experimental results indicate that accuracy of this method is superior in terms of traditional road lane extraction in complicated traffic condition.