In this paper, we present a real time visual tracking algorithm to detect and track a moving object. The corner points extraction algorithm detects corner points of the moving object, matches corner points and estimate corner point's location. The corner points are detected by arm method during image processing. The matching operation is performed at every corner point of the search region by normal correlation. Extended Kalman Filter(EKF) is used to predict search region whose size is based on the predicted position covariance in the new frame.
In the experiment, the EKF-based method successfully tracks the corner points of an object moving in the 2D space.
In 3D space, a triangle consisting of two corner points and the mass center of a moving object is used for calculating the distance and the surface normal of the object. By using 1st order Differential Invariants, visual information can be extracted from the moving object for a small field of view. Time-to-contact and the surface normal of the object can be determined by the triangle of object feature points. The proposed pose estimation method is investigated for estimating the pose of a moving object in 3D space. We obtained accurate results in the simulation for the case the Affine assumtion is satisfied. In the real experiment, however, the estimated pose of the object showed a relatively small error compared to the simulated one, due to the the error in the corner point's location.