For the purpose of locating pedestrian ahead of vehicle faster and more accurately, this paper presents a pedestrian detection method using boosted histograms of oriented gradients (HOG) features in the region of interest (ROI). These features are extracted in the regions where the pedestrian's legs may exist. Then the gentle AdaBoost learning algorithm is adopted to select some discriminative features and to form a strong cascaded classifier to identify pedestrian. The week classifiers are optimized by the weighted fisher linear discriminant (WFLD) combined with look up table (LUT) gentle AdaBoost instead of the linear SVM, which is helpful to accelerate the training and detection processes. Experimental results indicate that this method can address the computational problems and achieve a good accuracy.