Dense optical flow estimation for event cameras: application to parcel sorting conveyors.

Event Cameras respond to brightness changes in the scene asynchronously and independently for every pixel, providing an output that is called ''event''. This camera has many advantages which make it particularly suitable to estimate optical flow. The goal of this project is to estimate a dense optical flow that is pixel-dependent, starting from a set of events. We refer to the state-of-the-art, applying some changes and giving some more detailed mathematical reasoning. We focus on the Steepest Descent and the Accelerated Proximal Gradient algorithms, adding in the second case a Vectorial Total Variation regularizer. Furthermore, we test the method at first on some synthetic data and then on a real dataset generated by the movement of a parcel on a conveyor.