Technique to model the movement of the scene using image sequence

Abstract

The paper proposes a technique to model the movement of a scene based on backward stochastic gradient pixel-by-pixel estimation of the deformation field. Calculation of the moving object’s area is considered as the task of testing the hypothesis that the image grid nodes belong to the area of motion. The paper presents the estimated two kinds of errors: false positive and false negative. The obtained results are compared with the results of the MVFAST algorithm. The increase in the signal-to-noise ratio of the image of a moving object is achieved by combining the frames of the video sequence under study. To combine conjugate points, high-speed recurrent algorithms are used that do not require a priori information. The paper presents an example of estimation of object’s trajectory parameters using the technique. The interframe geometric deformations of the video sequence are used as intermediate parameters of the trajectory.