A Method of Robust Pedestrian Tracking in Video Sequences Based on Interest Point Description

Abstract

There are several methods proposed for detection and segmentation of object effectively. However, this algorithm struggles to detect an object with a lot of noise and shadows. Therefore, it is difficult to segment the accurate region and information of the object using background modeling only. To solve these problems, this paper introduced a more effective method of object segmentation based on interest point detection and description, which are core SURF theories. As a result, the feature extracted from the region of interest (ROI) was detecTable even with changes in scale, noise, and illumination. We then made the adaptive search window by this feature for ROI. After object detection, we applied the SVM to train the information of the feature from the detected object, and a classifier was built to estimate whether a result was a pedestrian. Therefore, if the result is a pedestrian, we would employ the Camshift algorithm to track the motion of this pedestrian. The experimental results showed the effectiveness of our method through comparison with others.

An Improvement of Pedestrian Detection Method with Multiple Resolutions

“…Even if they are locally overlapped, many algorithms have the capability to detect these targets [7]. However, in the case low resolution, the pedestrians which contain a small amount of information cannot be detected easily.…”