An Object Tracking Algorithm Based on Projective Invariants

An object tracking algorithm based on Planar Projective Invariants is proposed in this paper. In the algorithm, the invariants for object modelling and tracking are calculated with the lines extracted from images. In order to obtain the line edges in images, all edges are thinned to be one pixel wide at first by an improved Sequential Thinning algorithm. Then a fast computing method is used to estimate the curvatures of the edges, and the edge points at which the curvatures are less than some threshold are collected for line fittings. Lines are selected by a neighbour-rule or window-rule to compute the invariants. The experiments show that the preprocessing algorithm proposed in this paper works quite well in extracting lines from images. The invariants calculated with those lines are very stable from different viewpoints, and the tracking algorithm is robust and suitable for practical use.