Using quasi-continuous histograms for fuzzy main motion estimation in video sequence

This paper presents a new fuzzy framework for main motion estimation in video sequences. The estimation is performed using a fuzzy representation of pixel gray levels. The motion is characterized by a set of parameters such as horizontal translation, rotation, etc. The method is based on a Hough-like vote procedure. In this scheme, the parametric space is discretized and each pixel votes for each bin of this discrete space. The votes are accumulated in a “quasi-continuous histogram’’ (QCH). The use of possibility theory and imprecise probabilities provides an accurate estimation of the histogram's mode related to the main motion. The advantages of quasi-continuous histograms in terms of accuracy and robustness are discussed in this paper. Very promising results were obtained using real and simulated video sequences. Comparative studies with classical methods are also presented.