Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Artifact Detection in Challenging Conditions
Detecting artifacts in challenging environments using a multimodal sensor setup. The goal is to design robust artifact detection deep neural networks that can operate in challenging conditions with respect to illumination and other noise sources such as dust and smoke.
Keywords: Convolutional Neural Networks, Object Detection, Deep Learning, Multimodal Sensing
The advances in object detection using deep learning directly improve robots’ perception of their environment. Sometimes though, robots must operate in challenging environments which are not well represented in the training data, and therefore the data and methods must be adapted to fit these particular environments.
This project will focus on an underground exploration and artifact detection in difficult conditions, featuring changing or no external illumination, dust, smoke and other hazardous situations. In these cases simply using RGB images, that typical object detectors are trained on, is not enough. To target these challenges in this project we propose to use a multimodal sensor setup featuring among others laser pointclouds, RGB and infrared images. The goal of the project will be to combine the information from these sensors, even over short time frames to detect artifacts in an underground environment.
The advances in object detection using deep learning directly improve robots’ perception of their environment. Sometimes though, robots must operate in challenging environments which are not well represented in the training data, and therefore the data and methods must be adapted to fit these particular environments.
This project will focus on an underground exploration and artifact detection in difficult conditions, featuring changing or no external illumination, dust, smoke and other hazardous situations. In these cases simply using RGB images, that typical object detectors are trained on, is not enough. To target these challenges in this project we propose to use a multimodal sensor setup featuring among others laser pointclouds, RGB and infrared images. The goal of the project will be to combine the information from these sensors, even over short time frames to detect artifacts in an underground environment.
- Literature review on object detection and related fields
- Exploring the use of currently available object detection
- Adapting these methods to use input from multiple sensor sources
- Including temporal information into the prediction
- Evaluating against baseline solutions
- Literature review on object detection and related fields - Exploring the use of currently available object detection - Adapting these methods to use input from multiple sensor sources - Including temporal information into the prediction - Evaluating against baseline solutions
- Motivated and independent student
- Knowledge in Python
- Strong interest/background in machine learning
- Experience with ROS is beneficial
- Motivated and independent student - Knowledge in Python - Strong interest/background in machine learning - Experience with ROS is beneficial
If you are interested in this project, please send your transcripts and CV to
Andrei Cramariuc (andrei.cramariuc@mavt.ethz.ch) and Lukas Bernreiter (lukas.bernreiter@mavt.ethz.ch)
More details at https://docs.google.com/document/d/1BbnsHhPLJCKaaVlU5H1g2GiaBWGpW1XJYD7vZOHBws8/edit?usp=sharing
If you are interested in this project, please send your transcripts and CV to Andrei Cramariuc (andrei.cramariuc@mavt.ethz.ch) and Lukas Bernreiter (lukas.bernreiter@mavt.ethz.ch)
More details at https://docs.google.com/document/d/1BbnsHhPLJCKaaVlU5H1g2GiaBWGpW1XJYD7vZOHBws8/edit?usp=sharing