Fast Camera Motion Estimation utilizing Mode and Directional Self Information in the Compressed Domain

Abstract

This paper presents a fast camera motion estimation technique based on analyzing the magnitude and orientation histograms obtained from the compressed domain motion vectors. The magnitude and orientation histograms obtained from the compressed domain motion vectors are analyzed separately in order to determine the outlier thresholds. The statistical mode of the magnitude and orientation histograms combined with the Directional Self Information of the orientation histogram serves as the threshold selection criteria for the outliers based on which the inlier motion vectors are identified. The inlier motion vectors are then fed to the least square estimator which calculates the camera motion parameters. Comparative analysis is carried out with three existing state-of-the-art camera motion estimation methods to establish the proposed technique. Our experiments show that the proposed technique is computationally faster at the same time retaining its accuracy in determining the camera motion parameters.