Today I made a better camera mount and did more tests.

 

Camera mount

 

To get compareable results I need to place camera and temperature sensor always on the same spot above the cooktop. So I took a few wooden planks and made a new mounting arm for Raspberry Pi and camera. It places the camera 60 cm above the center of the cooktop.

 

Here is a photo from the front:

Camera mount above cooktop from the front

And here is a photo from top so you can imagine what the sensor sees:

Camera mount from top

The sensor is placed where normally the range hood sits. Maybe I should team up with aspork42 to get sensor and firmware into his Smart Range Hood - Pi Chef Challenge Blog post #1

 

Test measurements

 

Because I had the problems with the measurement in my last blog post I decided to dig deeper into this. To make easier reproduceable measurements I decided to simply boil water in a small pot. You can see the test drive in the following video.

 

 

Again you can see that the measured temperature only goes up to around 65 degrees celsius. But the water in the pot is definitely boiling because there are bubbles anywhere in the water. So the temperature should be around 100 degrees celsius.

 

So I started to figure out what is the reason for this behaviour.

 

Interpolation

 

First I used another OpenCV algorithm to scale the temperature data up to the camera image where I can see where the real grid-eye sensor pixels are and nothing interpolated.

Before I had this code:

cv::resize(outSmall,outSmallnorm,cv::Size(320,320));

now I use:

cv::resize(outSmall,outSmallnorm,cv::Size(320,320),0,0,cv::INTER_NEAREST);

 

This prevents interpolation and I could move the pot so that one pixel lies completely in the middle of the pot.

But it turned out that this was not the problem.

 

Out-of-center-error

 

Then I thought: maybe it is because the pot is not in the center of the image and it somehow gets confused with the angle.

So I moved the pot with the boiling water to the center.

 

It turned out that this was also not the problem. The temperature stayed at around 65 degrees celsius.

 

distance to sensor

 

Then I decided to lift the pot and bring the boiling water closer to the sensor. This helped and the temperature rose to about 100 degrees celsius.

 

You can see all these tests in the following video:

 

Conclusion

 

It seems that the grid-eye sensor needs more pixels which lay side by side to have the same temperature to get accurate temperature measurements in the center. I will figure out how to make a test setup to verify this.

 

What's next?

 

More testing ...

and start with the HMI.