Tensorial Lucas-Kanade: An Optical Flow Estimator Based on Tensorial Color Representation and Tensorial Algebra

Color data and optical flow estimation are important attributes to be processed in color image sequence. To deal with this, it is important to define methods for color information measure and strategies to solve the well-known Aperture Problem. This work proposes a new optical flow estimation approach for color image sequences called Tensorial Lucas-Kanade Technique. The technique proposed is based on tensorial color representation, tensorial morphological gradient and Lucas-Kanade optical flow estimation technique. Experimental results, comparing to ground truth optical flow by two different criteria, demonstrate the accuracy of its application in several sequences of synthetic color images. Four different tensorial dissimilarity measures were used to evaluate the technique. Comparing to the Lucas-Kanade's, tensorial technique had smaller average error in 94% of cases, with different dissimilarity measures and in 47% of cases, using the Frobenius norm.