Nighttime vehicle detection using soft cascade adaboost learning and KLT tracker

We propose an algorithm to detect vehicles in front of an ego-vehicle using the soft cascade AdaBoost algorithm and a KLT feature tracker-based tracking method for road traffic scene at night. The proposed algorithm enhances the previous algorithm [4] in terms of processing time and vehicle detection performance. The algorithm also improves the tracking performance of the detected vehicles between consecutive image frames even with abrupt changes in the pitch or yaw motion of the ego-vehicle. We also produce a learning data set by exploiting an active AdaBoost learning scheme, which further improves classifier performance. The experimental results for real road images illustrate the effectiveness of the proposed algorithm. © ICROS 2019.