Adaptive IPM-Based Lane Filtering for Night Forward Vehicle Detection

This paper presents a new vision-based vehicle detection method for Forward Collision Warning System (FCWS) at nighttime. Also, lane detection is performed for assistance. To effectively extract the bright objects of interest, an essential image preprocessing including the tone mapping, contrast enhancement and adaptive binaryzation is applied in the nighttime road scenes. The characteristics of taillights in gray-level image are extracted by night vehicle detection method, and the resulted taillight candidates are verified by their corresponding red-component which results from R and B color channels. The taillight candidates to be performed with pairing algorithm are filtered by our proposed adaptive lane boundaries on the basis of Inverse Perspective Mapping (IPM). In addition, we proposed a new detecting scheme which performs the detecting algorithm on two Region of Interest (ROI) defined by different size each time. The computing burden is then reduced because vehicle detection does not have to be performed on the entire image. Finally, relative distance and Time To Collision (TTC) are estimated to warn the inappropriate driving behavior of the driver. The proposed night vehicle detection which integrates lane detection has successfully implemented in ADI-BF561 600MHz dual-core DSP.