Abstract: In view of the traditional pedestrian detection which use single global feature whose expression ability is insufficient and keep out the pedestrian detection rate is low, we propose pedestrian detection algorithm in this paper based on the Fisher vector and the body parts model. Algorithm with Fisher vector quantify the HOG features of human body parts, and support vector machine (SVM) to learning for integral human body classifier and part body classifier. Then we use Hough vote on integral classifier and the classification of the part body classifiers to vote results, the highest score represents. Finally, we use Non-Maximum Suppression to eliminate false alarm. Through the testing in the standard pedestrian database and comparing to the current popular pedestrian detection algorithm, the experimental results show that the proposed method of multi-scale pedestrian detection in complex scene and pedestrian occlusion detection has good accuracy and robustness.