Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Learning-based analog gauge reading in the wild
Since time immemorial, humans have built instruments to monitor and measure the performance of their creations. Most such gauges are analog and meant to be read only by humans. For robots, reading analog instruments is a challenge, as they are typically small, come in many shapes and sizes, use various units and often have a reflective glass surface to protect the face of the gauge, which causes difficulties for robotic perception.
Giving robots the ability to read analog instrumentation would allow them to monitor and interact with legacy hardware that was not originally designed for robot interaction.
At the Autonomous Systems Lab, we are building a mobile manipulation robot that can inspect open world industrial environments and perform various routine maintenance tasks.
In this project, we will build an algorithm that can read analog pressure gauges.
We will investigate different methods and approaches for reading analog instrumentation, including, but not restricted to: optical character recognition, keypoint and object detection, semantic segmentation, shape detection, image warping and feature learning.
The goal would be to develop an algorithm, that given an image and bounding box detection of a pressure gauge, can estimate the current reading of the gauge and predict or extract the units used by the gauge.
The algorithm will be deployed in a real-world inspection and maintenance application.
This project is a unique opportunity to develop a novel algorithm of high practical relevance and impact, which will be deployed in a real world scenario. Not only that, but it will also be an algorithm and application that even your family and friends can understand the significance of.
For related work, see [1-6].
We encourage students to publish their work by releasing the code open-source and communicating their findings through a publication.
In this project, we will build an algorithm that can read analog pressure gauges.
We will investigate different methods and approaches for reading analog instrumentation, including, but not restricted to: optical character recognition, keypoint and object detection, semantic segmentation, shape detection, image warping and feature learning.
The goal would be to develop an algorithm, that given an image and bounding box detection of a pressure gauge, can estimate the current reading of the gauge and predict or extract the units used by the gauge.
The algorithm will be deployed in a real-world inspection and maintenance application.
This project is a unique opportunity to develop a novel algorithm of high practical relevance and impact, which will be deployed in a real world scenario. Not only that, but it will also be an algorithm and application that even your family and friends can understand the significance of.
For related work, see [1-6].
We encourage students to publish their work by releasing the code open-source and communicating their findings through a publication.
- Design of an algorithm that can read the value and units from an image of an analog pressure gauge
- Implementation of the algorithm
- Training and evaluating the algorithm on real-world data
- Annotated datasets will be provided
- Discovering the limits and failure modes of the algorithm
- Design of an algorithm that can read the value and units from an image of an analog pressure gauge - Implementation of the algorithm - Training and evaluating the algorithm on real-world data - Annotated datasets will be provided - Discovering the limits and failure modes of the algorithm
- Python or C/C++ programming experience
- Experience with Machine Learning
- Experience with git and SW development best practices is a plus
**Related Work**
[1] Charig Yang, Weidi Xie, & Andrew Zisserman (2021). It's About Time: Analog Clock Reading in the Wild. CVPR 2022.
[2] https://github.com/VictorSuarezVara/Reading-analog-clocks-with-neural-networks
[3] Howells, Ben, James Charles, and Roberto Cipolla. "Real-time analogue gauge transcription on mobile phone." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021.
[4] Laroca, Rayson, et al. "Towards image-based automatic meter reading in unconstrained scenarios: A robust and efficient approach." IEEE Access 9 (2021): 67569-67584.
[5] S. Dumberger, R. Edlinger and R. Froschauer, "Autonomous Real-Time Gauge Reading in an Industrial Environment," 2020 25th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA), 2020, pp. 1281-1284.
[6] Deng, G.; Huang, T.; Lin, B.; Liu, H.; Yang, R.; Jing, W. Automatic Meter Reading from UAV Inspection Photos in the Substation by Combining YOLOv5s and DeeplabV3+. Sensors 2022, 22, 7090.
- Python or C/C++ programming experience - Experience with Machine Learning - Experience with git and SW development best practices is a plus
**Related Work**
[1] Charig Yang, Weidi Xie, & Andrew Zisserman (2021). It's About Time: Analog Clock Reading in the Wild. CVPR 2022.
[3] Howells, Ben, James Charles, and Roberto Cipolla. "Real-time analogue gauge transcription on mobile phone." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021.
[4] Laroca, Rayson, et al. "Towards image-based automatic meter reading in unconstrained scenarios: A robust and efficient approach." IEEE Access 9 (2021): 67569-67584.
[5] S. Dumberger, R. Edlinger and R. Froschauer, "Autonomous Real-Time Gauge Reading in an Industrial Environment," 2020 25th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA), 2020, pp. 1281-1284.
[6] Deng, G.; Huang, T.; Lin, B.; Liu, H.; Yang, R.; Jing, W. Automatic Meter Reading from UAV Inspection Photos in the Substation by Combining YOLOv5s and DeeplabV3+. Sensors 2022, 22, 7090.
If interested, please contact us via email and include a brief explanation of your motivations for doing this project as well as some relevant prior work you did. Please also attach your CV and transcript of records.
Send the email to **all of the following recipients**: julian.keller@mavt.ethz.ch, victor.reijgwart@mavt.ethz.ch, kblomqvist@mavt.ethz.ch
**Do not apply directly via SiROP, but use the provided emails to get in touch with us.**
If interested, please contact us via email and include a brief explanation of your motivations for doing this project as well as some relevant prior work you did. Please also attach your CV and transcript of records. Send the email to **all of the following recipients**: julian.keller@mavt.ethz.ch, victor.reijgwart@mavt.ethz.ch, kblomqvist@mavt.ethz.ch
**Do not apply directly via SiROP, but use the provided emails to get in touch with us.**