Pedestrian Tracking Utilizing Scale Invariant Feature Transform and Particle Filter

Background: Pedestrians are the major road users in transportation system. They are more vulnerable than other road users when traffic accidents occurr, which has attracted much concerns from researchers around the world by developing corresponding countermeasures. Pedestrians are not easy to be tracked accurately because of the changes in illumination conditions and the occlusion of human body using traditional tracking algorithms. Method: To improve the effectiveness of pedestrian tracking, particle filter (PF) is utilized to track the pedestrian, which is detected using the histograms of oriented gradient (HOG) features. Then scale invariant feature transform (SIFT) features are employed to represent the region of interest for sequence images. Result: The representative vector utilized to describe the pedestrian is renewed after comparing the object model and the characteristic variables during the tracking process. This method takes advantage of color histogram and adopts PF to predict the position of the pedestrian. Conclusion: Experiments were conducted to compare the proposed method with traditional PF tracking method. Results verify the accuracy and efficiency of the proposed method.