Description

In recent years, unsupervised deep learning approaches have received significant attention to estimate the pixels' depths and vehicle odometry from unlabelled monocular image sequences. That is, predicting the six-degrees-of-freedom pose camera motion and depth map of the scene from unlabelled RGB image sequences.
Deep Learning, in turn, excels in solving image data association problems such as object detection, segmentation or classification.

This MSc thesis proposes to combine the camera's ego-motion and map estimation with other image analysis techniques to provide a local robo-centric map extended with traversability properties such as object classification or semantic understanding.

Requirements:
- Degree: enrolled in an MSc degree related to Computer Science, Informatics, Telecommunications, Industrial Engineering or Mathematics.
- Background: Experience in machine learning and computer vision. Good programming skills with C++ and ROS.
- Language: Fluent in English