Hi All,
I’m a high school student and a beginner in machine learning. I’ve previously worked on a basic project using Edge Impulse and an ESP32 for gesture recognition. Now, I’m working on a STEM project involving a breath sensor to analyze and evaluate breathing patterns.
The sensor is a variable resistor connected to an ESP32, and its resistance changes based on the user’s breathing:
- Breath In: Resistance decreases (e.g., from ~1000 to ~800).
- Hold: Resistance stabilizes (e.g., around ~800).
- Breath Out: Resistance increases (e.g., back to ~1000).
The challenge I’m facing is that the resistance values are noisy and don’t stay within fixed thresholds, as the baseline values and their variations fluctuate over time.
My goal is to:
- Detect the breathing phases (inhale, hold, exhale) dynamically based on the sensor data.
- Allow the user to preselect a breathing pattern (e.g., 3 seconds inhale, 2 seconds hold, 4 seconds exhale) and analyze their breathing session over 5 minutes.
- Calculate the percentage deviation from the preselected pattern to evaluate how closely the user followed it.
I’d love your guidance on:
- What would be the best approach to model this using Edge Impulse?
- How can I handle the noise and dynamic threshold variation effectively in this project?
Thank you so much for your help! Any advice or suggestions are greatly appreciated!