A method for visualizing pedestrian traffic flow using SIFT feature point tracking

Abstract

This paper presents a method for visualizing a pedestrian traffic flow using results of feature point tracking. The Kanade-Lucas-Tomasi feature tracker algorithm for point feature tracking is widely used because it is fast; however, it is sometimes fails to accurately track non-rigid objects such as pedestrians. We have developod a method of point feature tracking using a scale invariant feature transform (SIFT). Our approach uses mean-shift searching to track a point based on the information obtained by a SIFT. We augment the mean-shift tracker by using two interleaved mean-shift procedures to track the mode in image and scale spaces, which represents the spatial location and the scale parameter of the keypoint, respectively. Since a SIFT feature is invariant to changes caused by rotation, scaling, and illumination, we can obtain a beter tracking performance than that of a conventional approach. Using the trajectory of the points obtained by our method, it is possible to visualize traffic pedestrian traffic flow using the location and scale obtained by SIFT feature point tracking. © 2009 Springer Berlin Heidelberg.