Water Level Measurement From Images Using Object Detection
Näs, Bo-Anders (2023)
Näs, Bo-Anders
2023
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2023102728065
https://urn.fi/URN:NBN:fi:amk-2023102728065
Tiivistelmä
The technology is advancing rapidly and today object detection is becoming increasingly common. Using a camera and object detection in water level measurements is useful and affordable, as cameras are often readily available at such locations. In this thesis, an application was developed to measure water level from images of a staff gauge using a Yolov5m model.
The quality of the captured images is often a problem. Strong light, reflections, and dirty staff gauges often cause the object detection to fail. To this category belongs also night-time and blurry images. A separate application was developed to handle this filtering. This application filters out these night-time and blurred images, images that otherwise the Yolov5m model would fail on.
In this thesis, two alternatives for the water level measuring application were compared. In the first alternative Yolov5m was used for detecting the staff gauge and numbers on the staff gauge and image processing was used for locating the water line. In the second alternative, the Yolov5m model was used for all three detections, for the staff gauge detection, the numbers detection on the staff gauge, and for the water line detection on the staff gauge. The second alternative using the Yolov5m model for water line detection outperformed the first alternative. The second alternative estimated the water level correctly in 40% of the test images and 83% of the test images, it estimated the water level within 3 cm of the correct level.
The quality of the captured images is often a problem. Strong light, reflections, and dirty staff gauges often cause the object detection to fail. To this category belongs also night-time and blurry images. A separate application was developed to handle this filtering. This application filters out these night-time and blurred images, images that otherwise the Yolov5m model would fail on.
In this thesis, two alternatives for the water level measuring application were compared. In the first alternative Yolov5m was used for detecting the staff gauge and numbers on the staff gauge and image processing was used for locating the water line. In the second alternative, the Yolov5m model was used for all three detections, for the staff gauge detection, the numbers detection on the staff gauge, and for the water line detection on the staff gauge. The second alternative using the Yolov5m model for water line detection outperformed the first alternative. The second alternative estimated the water level correctly in 40% of the test images and 83% of the test images, it estimated the water level within 3 cm of the correct level.