Tez

Permanent URI for this collectionhttp://acikerisim.bau.edu.tr:4000/handle/123456789/160

Browse

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Inertial sensor fusion for 3D camera tracking
    (Bahçeşehir Üniversitesi Fen Bilimleri Enstitüsü, 2012-02) Özer, Nuri; Eroğlu Erdem, Çiğdem
    3D motion tracking becomes more important in computer vision with increase of robotics and augmented reality's (AR) applicable areas such as medical education, remote robot control, entertainment and cultural heritage. In order to achieve a realistic feeling of immersion, the rendering of the virtual content has to be in alignment with real objects in the video and this requires a high-accuracy 3D tracking. The methods using only camera measurements generally perform well at slow camera motion; however they become less accurate at high velocities and accelerations due to motion blur. Inertial sensors on the other hand measure the derivatives of the camera pose and hence can be employed to improve the tracking performance at high velocities and accelerations, but cannot perform well at slow motion because of the error drift. Therefore, we present a high-accuracy 3D camera tracking method using inertial sensors but not require placing any devices or points on the scene. 3D information of scene where 3D motion tracking is done is previously known. The method consists of an Extended Kalman filter (EKF) that fuses the information from visual and inertial sensors. A hybrid filter combining the Bayesian filter and the direct linear transformation (DLT) is also used instead of EKF. The biases of the inertial sensors are also considered during the motion. In addition to performance comparison of these two filter, the performance of using both or one of accelerometer and gyroscope measurements as control input is compared to using both or one of accelerometer and gyroscope measurements as measurement. It is concluded via simulations that using inertial sensors in 3D camera tracking gives more accurate results and using inertial sensors as measurement or control input does not affect the performance of 3D camera tracking, while providing a lower complexity tracker. Also, EKF always performs better than the hybrid filter in simulations.