Comparative Analysis of Dynamic Object Segmentation Networks for Tele-robotic Applications

A telerobotic system exists which captures images of a remote environment and builds a map in VR of the aforementioned environment which can be viewed at a different location. A significant problem which this system faces is the proper tracking of dynamic objects in SLAM. When it came to dynamic objects, suitable updating of the map was not successful as a result of poor tracking of the images.

The main goal of this assignment is to develop a module that will enable proper tracking of dynamic objects by segmentation in SLAM to update the map in VR efficiently without any errors, giving a replica of the original remote environment. Several techniques have been used over the past decade for segmentation and tracking of dynamic objects such as Background Subtraction, Optical flow algorithms, deep learning-based methods, and so on.

In this assignment, one or a combination of techniques will be used to enable the proper segmentation of dynamic objects.