Configurable pointer meter reader based on computer vision
Järvinen, Niko (2022)
2022
Master's Programme in Computing Sciences
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2022-05-25
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202205185052
https://urn.fi/URN:NBN:fi:tuni-202205185052
Tiivistelmä
With the global demand for industrial and home automation growing, new solutions that utilize existing installations are needed. Neural network-based solutions have been introduced to the field, but they lack configurability. Unlike neural networks, traditional computer vision algorithms can trivially be made configurable, but many studies make significant assumptions and delimitations regarding different types of meters. This Master's thesis aimed to find a configurable, accurate, and performant combination of algorithms for each phase of the meter reading and discuss how the proposed system could be improved in terms of performance and robustness.
A test bench with five distinct meters was built in order to test the system thoroughly. Numerous algorithms used by other authors were inspected and compared. Various algorithms not utilized in the proposed system were included in order to discuss their advantages and disadvantages.
This Master's thesis proposes a configurable, accurate, and performant pointer meter reading system. The proposed system utilizes scale-invariant feature transform (SIFT), inverse perspective mapping, a new color segmentation approach, annotated templates, binary mask thinning, and random sample consensus (RANSAC). The new color segmentation algorithm is fast, accurate, and intuitively configurable and has already been utilized in industrial deployment prior to publishing this thesis.
The system is tested by reading five meters simultaneously from various angles. Regardless of the rigorous experimental setup, the results show that the system is on par with the meter manufacturers' accuracy guarantees and often exceeds them significantly.
A test bench with five distinct meters was built in order to test the system thoroughly. Numerous algorithms used by other authors were inspected and compared. Various algorithms not utilized in the proposed system were included in order to discuss their advantages and disadvantages.
This Master's thesis proposes a configurable, accurate, and performant pointer meter reading system. The proposed system utilizes scale-invariant feature transform (SIFT), inverse perspective mapping, a new color segmentation approach, annotated templates, binary mask thinning, and random sample consensus (RANSAC). The new color segmentation algorithm is fast, accurate, and intuitively configurable and has already been utilized in industrial deployment prior to publishing this thesis.
The system is tested by reading five meters simultaneously from various angles. Regardless of the rigorous experimental setup, the results show that the system is on par with the meter manufacturers' accuracy guarantees and often exceeds them significantly.