Camera Pose Estimation in Dynamic Scenes with Background Tracking

• Main Authors: Dong ZHANG, Ping LI
• Format: Article
• Language: English
• Published: IFSA Publishing, S.L. 2013-02-01
• Series: Sensors & Transducers
• Subjects: Camera pose estimation; Motion detection; Multiple view geometry; Dynamic vision
• Online Access: http://www.sensorsportal.com/HTML/DIGEST/february_2013/P_SI_310.pdf

This paper presents a method for robustly estimating camera pose in dynamic scenes. The main contribution of it is to use a background tracking method to ensure the accurately tracking of the moving background, thus to increase the robustness of the camera pose estimation. It has three innovations over existing methods. First, it uses a fast model initialization method to get the sparse model of the background; Second, it utilizes the newly developed 3-D motion segmentation method to segment the consecutive frames into different motions and uses an on-line model updating method to track the sparse model of the background; Third, it employs a model selection method to effectively connect the dynamic scenes with static scenes for an accurate estimation result. The method is tested on synthetic data and real-world image sequences and results show that it outperforms the existing methods.