Video Object Segmentation using Optical Flow and Recurrent Neural Networks

Abstract

This paper implements the task of semi-supervised Video Object Segmentation (VOS), i.e., the separation of an object from the background in a video, given the mask of the first frame. To accomplish this task, modern Machine learning techniques have been used, such as, Convolution Neural Networks (CNNs) and Convolutional Recurrent Neural Networks (CRNNs). The motion of objects between consecutive frames of the sequence, caused by the relative movement between the object and camera is very important information in this task and bears the name Optical Flow (OF). OF was used in the paper to improve the algorithm performance. Experiments, carried out on the Densely Annotated VIdeo Segmentation (DAVIS) database, show competitive results as other algorithms from this field.