You mentioned about training a model for thermal image recognition, but in the end this project is to monitor the “emotion health”. Any plan on collecting dataset for this purpose, and what kind of machine learning method you’re going to use?
Our initial idea is that the temperature reading alone would supplement the RGB image result. If the result from RGB image is ‘angry’, then the temperature would likely be higher, as King Hei mentioned. If we are taking an approach purely using thermal image, we need to produce our own dataset thermal images with emotion labels, since we could not find any online. We chose not to do this, as we think it is not feasible to create this dataset along with all our other tasks.
MOW, Wai Ho
January 19, 2022 2:35 pm
Why choose the thermal cam sensor MLX80640 instead of other brands like FLIR or SEEK ones?
When we are doing the research, we try to find the thermal camera for raspberry pi on the internet. Many websites recommend Thermal camera MLX90640 as it is a cheap and high-resolution camera.
In order to match our project objectives, we choose this low-cost and effective thermal camera.
We think the same. Our plan is to simply use the temperature reading as an input to the system. We will likely also revert to using the FLIR cams for thermal, as they have been used successfully in previous projects.
You mentioned about training a model for thermal image recognition, but in the end this project is to monitor the “emotion health”. Any plan on collecting dataset for this purpose, and what kind of machine learning method you’re going to use?
Good question! The team should explain how the thermal cam is used for emotion detection.
We are still thinking about it, maybe we will mostly focus on the temperature. For example, if a person is angry, his temperature should be higher.
Our initial idea is that the temperature reading alone would supplement the RGB image result. If the result from RGB image is ‘angry’, then the temperature would likely be higher, as King Hei mentioned. If we are taking an approach purely using thermal image, we need to produce our own dataset thermal images with emotion labels, since we could not find any online. We chose not to do this, as we think it is not feasible to create this dataset along with all our other tasks.
Why choose the thermal cam sensor MLX80640 instead of other brands like FLIR or SEEK ones?
When we are doing the research, we try to find the thermal camera for raspberry pi on the internet. Many websites recommend Thermal camera MLX90640 as it is a cheap and high-resolution camera.
In order to match our project objectives, we choose this low-cost and effective thermal camera.
How much is it? What’s its resolution and frame rate?
Around 500 hkd. Resolution is 32×24 pixels, frame rate is 16Hz.
Cheap but not that high resolution. What is the max distance that one can still take temperature from the kit?
The resolution is also my concern. Is such low resolution sufficient for this application?
Guess not. 🙂
We think the same. Our plan is to simply use the temperature reading as an input to the system. We will likely also revert to using the FLIR cams for thermal, as they have been used successfully in previous projects.
up to 7m
Must be very inaccurate (like +/10degrees) at 7m.
We will change to use the FLIR thermal camera.