This is a python application which takes an image or video frame of an analog gauge as an input and reads the value using the OpenCV Computer Vision library.
To convert the calculated angels of the detected lines we need some reference angels and their associated values. You can find the example images used to calibrate our gauge reader in the gauge_images directory.
Example Gauge showing ~1.25 bar
In the following table you can see the detected angels and the respective estimated value of the gauge:
| detected angle | estimated value (bar) |
|---|---|
| -15° | 1.25 |
| -32° | 1 |
| 17° | 2 |
| -39° | 0.8 |
| 12° | 1.5 |
Using a affine / linear curve fitting equation finding method we can estimate the following mathematic equation from it:
f(x) ≈ 0.0179735x + 1.5249
Plot corresponding fitting curve
See https://www.dcode.fr/function-equation-finder for more information on how to find a affine / linear curve fitting equation
First, we perform edge detection with the Canny edge detector algorithm implementation from OpenCV.
After that we can use OpenCV's implementation of the probabilistic Hough transform algorithm HoughLinesP to detect lines in the previosly processed image.
This will then look like this:
At first glance, it looks like there are two clear lines on the image. Only at a closer look you can see that they are four lines, which were recognized by the alrogithm. For a better representation, numbers are attached to the respective lines in the following picture.
For measurement of the angle of the lines, we use the atan2 function, which is an extended version of the trigonometric inverse tangent function.
With that function we calculate the angles between all the vectors that make up the lines and take the average of them.
This value is then substituted into our equation described above (f(x) ≈ 0.0179735x + 1.5249) to obtain the desired pressure value in bar.
In the image above, the algorithm detects an average angle of -15°. If we put this into our equation we will end up with the following:
f(x) ≈ 0.0179735x + 1.5249
f(x) = 0.0179735 * -15 + 1.5249
f(x) = 1.2552975 bar
As we can see in the example gauge image, this value is almost 100% accurate.
- Prometheus Integration to show values in Grafana Dashboards and add alerts for certain tresholds
- Add support for periodically checking camera / video input
See https://docs.opencv.org/3.4/d9/db0/tutorial_hough_lines.html for more information about the OpenCV Hough Line Transform.