The Efficient Features for Tracking

In computer vision, optical flow is very useful in tracking objects from frame to frame. The well-known KLT (Kanade-Lucas-Tomasi) of the differential method can trace local features through a generalized affine transform explained by six parameters. Since most of the distortion of the local patch between frames can be explained by four factors (horizontal and vertical translation, scaling and rotation), we present a tracking method with four parameters in this paper. The advantages of our method are that it has lower complexity than KLT and calculates the values of the four parameters directly. In this experiment, we show the four estimated factors in a virtual scene and track local features in a real scene. As a result, despite reducing the parameters, the proposed method performs very well for feature tracking and can track local features using only four parameters.