Coralie Camus, "Extraction of optical flow fields from from a "silicon retina" optical sensor", 2010.

Traditional vision sensors (cameras) present some limits like a high latency that the dynamic vision sensor from ETH Zurich try to minimize. This project aims to extract motion vectors from events provided by the DVS128 in an asynchronous stream of data. In contrast to traditional computation using a xed time distance and estimating the spatial distances, the spatial distance are here xed and the time intervals estimated. For this, several Reichardt detectors are used and the evidence for a velocity is computed. The velocity with the maximum correlation value is the most likely. This report also explains how to improve the condence values to obtain a higher quality of motion vectors.