Estimación del movimiento de objetos en secuencias de imágenes de intensidad y profundidad

Abstract

The aim of this research work is to perform a system that uses optical flow for detecting and monitoring one or more objects in a sequence of images, including depth ones,to design an application that can differentiate between the moving objects in the sequence and that can follow their trajectories. In this sense, we uses an implementation of a well-known optical flow method that uses isotropic TV-L1 regularization and two variants of this approximation. These variants arise to improve the rounded effects that typically produces this smoothing strategy. Thus, we have add, in the original implementation, a decreasing function in the regularization term to inhibit the diffusion at high image gradients, where the flow discontinuities are located. This second approach presents a good contours definition that mitigate the rounded effects. However,it may produce some instabilities in the optical flow. To cope with this situation, a third implementation has been made to ameliorate this problem using a constant value in the decreasing function that assures a minimum isotropic diffusion to prevent these instabilities. On the other hand, once studied the displacement fields reached by depth images and seen their results, it has been developed a program that creates a new type of image that mixes the depth information and its corresponding mask. This creates a new image richer in details than the original. Thus, the optical flow calculation is favored. Building on the accuracy of these methods, it has been implemented different applications: (i) a system that generate videos matching the original sequence with its respective flow field (ii) a tracking system that follows an area selected by the user (iii) a more complex tracking system that follows different areas in the scene, (iv) a system to segment the moving objects in the scene (v) a system that mixes the depth images and their mask information.