This paper proposes a new real-time side pedestrian detection method for driver assistance systems. The approach includes two steps, classification and verification. In classification step, we use the Haar feature and a cascade of stage classifiers trained by Adaboost algorithm. In verification step, we use HOG feature and SVM classifier to refine false positives. To speed up the processing time, we use OCS algorithm in classification step. Meanwhile, we also propose a new dataset using an onboard camera to shoot side pedestrians on zebra in urban environment. For a 704*576 image, the system can process more than 15FPS. We experimentally found that the detection rate is about 90% while the false alarm rate is less than 1%.