909 FPGA Face Detection
909 : FPGA Face Detection
- Author: Nikita
- Description: Streaming pixel-based skin-tone and facial feature detector with eye, nose and mouth dark-pixel counting.
- GitHub repository
- Open in 3D viewer
- Open in Tiny Tapeout Explorer
- Clock: 50000000 Hz
How it works
Pixels are streamed into the chip serially, one channel at a time (R, G, B) over ui_in. The uio_in pins carry the channel select, pixel_valid flag, frame_done flag, and zone metadata (which third of the face the pixel belongs to).
The chip classifies each pixel for skin tone using RGB thresholds, and accumulates dark pixel counts in the eye, nose, and mouth zones. At end of frame (frame_done pulse), it evaluates all conditions and latches the result onto uo_out.
How to test
Use the MATLAB preprocessing script to convert a BMP image into a pixel stream CSV with zone metadata. Feed pixels via a microcontroller or the cocotb testbench. Pulse frame_done after the last pixel. Read face_detected on uo_out[0].
Inputs
| Pin | Description |
|---|---|
| ui_in[7:0] | 8-bit pixel channel value (R, G, or B) |
| uio_in[1:0] | Channel select: 00=R, 01=G, 10=B |
| uio_in[2] | pixel_valid |
| uio_in[3] | frame_done |
| uio_in[5:4] | row_zone: 00=eye, 01=nose, 10=mouth |
| uio_in[7:6] | col_zone: 00=left, 01=right, 10=centre |
Outputs
| Pin | Description |
|---|---|
| uo_out[0] | face_detected |
| uo_out[1] | skin_found |
| uo_out[2] | eyes_found |
| uo_out[3] | nose_found |
| uo_out[4] | mouth_found |
IO
| # | Input | Output | Bidirectional |
|---|---|---|---|
| 0 | pixel channel value bit 0 | face_detected | channel select bit 0 |
| 1 | pixel channel value bit 1 | skin_found | channel select bit 1 |
| 2 | pixel channel value bit 2 | eyes_found | pixel_valid |
| 3 | pixel channel value bit 3 | nose_found | frame_done |
| 4 | pixel channel value bit 4 | mouth_found | row_zone bit 0 |
| 5 | pixel channel value bit 5 | row_zone bit 1 | |
| 6 | pixel channel value bit 6 | col_zone bit 0 | |
| 7 | pixel channel value bit 7 | col_zone bit 1 |