The research of pedestrian detection ahead of vehicle is the front direction in the field of vehicle safety assistant driving at present. The method of SVM pedestrian detection based on HOG features is studied in this paper. Firstly, the histograms of oriented gradient features between pedestrian and non-pedestrian samples are extracted. Then the features are used as an input vector of SVM algorithm, getting pedestrian classifier with a higher recognition by training. Finally the trained classifier is loaded into the online pedestrian detection system to detect the transport road image. The experimental results show that the algorithm can effectively identify the different scales and attitude pedestrian in complex background.