Chapter 6: Color Recognition
Chapter goal: Have the NanoCam recognize the color of objects in the image, and get the coordinates for applications such as sorting.
Principle
Based on the HSV (Hue-Saturation-Value) color space. The RGB565 image output by the camera is processed by esp-dl's ColorDetector engine: the image is scaled down to 80×80 resolution to reduce noise, then converted pixel by pixel into HSV values, which are matched against 7 preset color thresholds.
Preset Color Thresholds (OpenCV standard H range, 0-180 scale)
| Color | Hue (H) | Saturation (S) | Value (V) | Area threshold |
|---|---|---|---|---|
| Red | 0-15 | 70-255 | 90-255 | 64 |
| Yellow | 23-33 | 70-255 | 90-255 | 64 |
| Green | 34-75 | 70-255 | 90-255 | 64 |
| Blue | 97-124 | 70-255 | 90-255 | 64 |
| Purple | 125-155 | 70-255 | 90-255 | 64 |
| White | 0-180 | 0-40 | 200-255 | 80 |
| Black | 0-180 | 0-255 | 0-50 | 80 |
Hue uses the OpenCV 0-180 scale (corresponding to 0-360°).
set_bgr(false)ensures the library reads the RGB565 data as-is without swapping channels.
Steps
6.1 Entering Color Mode
ai_mode:3The device reboots automatically into color detection mode, and the WS2812 RGB LED (GPIO18) displays the currently recognized color.
For the complete commands, see the Serial Protocol Manual.
6.2 Observing the Recognition Result
Put a solid-colored object in front of the camera; open http://<IP> in a browser and you will see:
A colored rectangle marking the detected color region
A color label text (red/yellow/green/blue/purple/white/black)
The box and label colors match the actually detected color
Color mode only does image overlay (OSD) and does not output serial logs. To get the coordinates, read them through the I2C registers.
6.3 Reading Detection Data over I2C
The NanoCam acts as an I2C slave (address 0x33, GPIO SDA=41 SCL=42) and updates the detection box center coordinates in real time.
| Register | Content | Data type |
|---|---|---|
| 0x28-0x29 | Center X | int16 BE |
| 0x2A-0x2B | Center Y | int16 BE |
| 0x2C-0x2D | Recognition ID | int16 BE |
Code
Core Detection Engine
components/modules/ai/who_color_detection.cpp — based on esp-dl ColorDetector:
// Build the detector; set_bgr(false) keeps the color channels correct
ColorDetector detector;
detector.set_bgr(false);
detector.set_detection_shape({80, 80, 1});
// Register the 7 color thresholds
detector.register_color({h_lo, h_hi, s_lo, s_hi, v_lo, v_hi}, area_min, "red");
// Detect
auto &results = detector.detect((uint16_t *)frame->buf,
{(int)frame->height, (int)frame->width, 3});
// Iterate over the results, drawing boxes + labels
for (int ci = 0; ci < (int)results.size(); ci++) {
for (int ri = 0; ri < (int)results[ci].size(); ri++) {
color_detect_result_t &res = results[ci][ri];
draw_rect(frame, res.box[0], res.box[1], res.box[2], res.box[3], color_lcd);
fb_gfx_print(frame, lx, ly, color_lcd, color_name);
}
}Result
Red/green/blue objects → color recognized → box + label drawn → coordinates output over I2C → can be connected to a servo for sorting.
Next chapter: Chapter 7: QR Code Scanning