A video stabilization algorithm for train-mounted intrusion detection system based on ALP keypoints

摘要

With the rapid increasing of railway mileage, automatic railway intrusion detection by analyzing the video from train-mounted camera is becoming very meaningful. But there exits serious jitter in the video since the camera always vibrates with train when it is running, a video stabilization reprocessing procedure is the prerequisite before the intrusion detection analysis. To solve this question, a robust video stabilization algorithm based on low degree polynomial detector is presented in this paper. Generally, this algorithm consists of three stages. Firstly, keypoints are extracted and tracked by ALP detector and KLT tracker respectively between adjoining frame of the video. Secondly, the local motion vector is obtained by image affine transformation model. Finally, the global motion vectors are corrected and smoothed by using RANSAC and Kalman filter respectively. At the end of this paper, experiments based on PSNR and ITF values are given to show that the proposed algorithm is more effective than Harris corner detector which is often used in the traditional stabilization algorithm.