Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Real-Time Semantic Segmentation for Aerial Vehicles
The goal of this project is to segment images that will be used for 3D reconstruction. This project is focused on segmenting parts of the image that are not interesting for the 3D reconstruction of buildings, such as, sky, ground and vegetation.
Keywords: Semantic Segmentation; 3D Reconstruction; Autonomous Navigation; Aerial Perception
Small, multi-rotor Unmanned Aerial Vehicles (UAVs) have become some of the most widely used robotic platforms due to their low cost and remarkable mobility. Nonetheless, their payload is very limited and so is the processing power and the sensing equipment that can be carried onboard. With these restrictions in mind, monocular cameras are some of the most appealing sensors for such vehicles, which can then be also used in visual-inertial Simultaneous Localization And Mapping (SLAM) for egomotion estimation and mapping the robot’s surroundings.
This project aims to semantic segment the SLAM map that is created onboard the UAV as fast as possible. This project is focused on segment parts of the image that are not interesting for the 3D reconstruction of buildings, such as sky, ground and vegetation.
Refs
Incremental Dense Semantic Stereo Fusion for Large-Scale Semantic Scene Reconstruction - (ICRA) 2015
https://youtu.be/k879Rw6BFy8
Small, multi-rotor Unmanned Aerial Vehicles (UAVs) have become some of the most widely used robotic platforms due to their low cost and remarkable mobility. Nonetheless, their payload is very limited and so is the processing power and the sensing equipment that can be carried onboard. With these restrictions in mind, monocular cameras are some of the most appealing sensors for such vehicles, which can then be also used in visual-inertial Simultaneous Localization And Mapping (SLAM) for egomotion estimation and mapping the robot’s surroundings.
This project aims to semantic segment the SLAM map that is created onboard the UAV as fast as possible. This project is focused on segment parts of the image that are not interesting for the 3D reconstruction of buildings, such as sky, ground and vegetation.
Refs
Incremental Dense Semantic Stereo Fusion for Large-Scale Semantic Scene Reconstruction - (ICRA) 2015 https://youtu.be/k879Rw6BFy8
- WP1: Research into state-of-the-art segmentation algorithms with low computational cost.
- WP2: Familiarisation with existing software for aerial SLAM.
- WP3: Development and integration of your semantic segmentation pipeline with the SLAM output.
- WP4: Experimentation and evaluation of these algorithms against state-of-the-art approaches.
- WP5: Further optimization of the pipeline of WP3 to work in challenging scenarios.
- WP6: Final evaluation of the methods and report writing.
- WP1: Research into state-of-the-art segmentation algorithms with low computational cost. - WP2: Familiarisation with existing software for aerial SLAM. - WP3: Development and integration of your semantic segmentation pipeline with the SLAM output. - WP4: Experimentation and evaluation of these algorithms against state-of-the-art approaches. - WP5: Further optimization of the pipeline of WP3 to work in challenging scenarios. - WP6: Final evaluation of the methods and report writing.
- Strong interest in computer vision.
- Strong analytical skills.
- Background in C++.
- Strong interest in computer vision. - Strong analytical skills. - Background in C++.
Your application must contain your most recent transcripts from bachelor and master studies. Lucas Teixeira - lteixeira@mavt.ethz.ch
Your application must contain your most recent transcripts from bachelor and master studies. Lucas Teixeira - lteixeira@mavt.ethz.ch