Published July 28, 2025

Smart Occupancy Detector with SCD41 CO₂ Sensor and ML

A compact ML-powered device that detects room occupancy using CO₂ and temperature data, ideal for smart environments.

BeginnerProtip2 hours149

Smart Occupancy Detector with SCD41 CO₂ Sensor and ML

Things used in this project

Hardware components

Arduino tiny ML

Sensirion SEK-SCD41

Jumper wires (generic)

Software apps and online services

Edge Impulse Studio

Arduino IDE

Story

In a world increasingly driven by smart solutions, one compact system brings human presence detection to a new level of precision. Born from the synergy of environmental sensing and machine learning, this Edge Impulse–powered project uses a Sensirion SCD41 CO₂ sensor and an Arduino TinyML board to estimate occupancy based on air quality.

As people enter or leave a room, CO₂ levels subtly shift. The trained machine learning model, developed and deployed using Edge Impulse, reads these fluctuations and classifies occupancy levels, zero, one, or multiple individuals without the need for intrusive cameras or motion sensors. Everything is packed into a portable and elegant configuration, powered by a few wires and intelligent code.

This setup isn’t just a demo. It’s a vision for energy-efficient buildings, responsive HVAC systems, and safer shared spaces. It shows how embedded ML can redefine what’s possible quietly, efficiently, and intelligently.

Code

Reading SCD41 Output and showing ML results on serial monitor

#include <Wire.h>
#include <SensirionI2cScd4x.h>
#include <Occupancy_Detector_inferencing.h>

SensirionI2cScd4x scd4x;
float features[EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE] = {0};

static bool debug_nn = false;

void setup() {
    Serial.begin(115200);
    while (!Serial);

    Serial.println("Starting occupancy detection with SCD41...");

    Wire.begin();
    scd4x.begin(Wire, 0x62);  // SCD41 I2C address is usually 0x62

    // Stop any previous measurement
    scd4x.stopPeriodicMeasurement();
    delay(500);

    // Start new measurement
    int16_t error = scd4x.startPeriodicMeasurement();
    if (error) {
        Serial.print("Failed to start periodic measurement: ");
        Serial.println(error);
    } else {
        Serial.println("Measurement started. Waiting for first reading...");
    }

    // Give it some time to warm up and produce valid data
    delay(5000);
}

void loop() {
    bool dataReady = false;
    if (scd4x.getDataReadyStatus(dataReady) != 0 || !dataReady) {
        Serial.println("Waiting for data...");
        delay(1000);
        return;
    }

    // Read sensor values
    uint16_t co2;
    float temperature, humidity;
    int16_t error = scd4x.readMeasurement(co2, temperature, humidity);

    if (error) {
        Serial.print("Error reading measurement: ");
        Serial.println(error);
        delay(1000);
        return;
    }

    Serial.print("CO2: ");
    Serial.print(co2);
    Serial.print(" ppm, Temp: ");
    Serial.print(temperature);
    Serial.println(" °C");

    // Format features for the model
    features[0] = (float)co2;
    features[1] = temperature;

    // Wrap data as Edge Impulse signal
    signal_t signal;
    int create_signal = numpy::signal_from_buffer(features, EI_CLASSIFIER_DSP_INPUT_FRAME_SIZE, &signal);
    if (create_signal != 0) {
        Serial.println("Failed to create signal from features!");
        delay(1000);
        return;
    }

    // Run classification
    ei_impulse_result_t result = { 0 };
    EI_IMPULSE_ERROR infer_result = run_classifier(&signal, &result, debug_nn);

    if (infer_result != EI_IMPULSE_OK) {
        Serial.print("Classifier failed: ");
        Serial.println((int)infer_result);
        delay(1000);
        return;
    }

    // Print inference result
    Serial.println("Predictions:");
    for (size_t ix = 0; ix < EI_CLASSIFIER_LABEL_COUNT; ix++) {
        ei_printf("  %s: %.5f\n", result.classification[ix].label, result.classification[ix].value);
    }

    delay(1000);  // Optional: Wait 1s before next inference
}