Seminars & Reports
ReportMay 10, 2025

ESP32-CAM Vision Report

Streaming, face detection and edge AI on a $10 board.

ESP32-CAM Vision Report

01What's on the board

ESP32-CAM combines the ESP32-S module with an OV2640 2 MP camera, a microSD slot and a small flash LED — all for around $10. No USB on board; you flash it with an external FTDI adapter.

02First flash

  • Wire FTDI: RX→U0T, TX→U0R, GND→GND, 5V→5V.
  • Short IO0 to GND to enter flashing mode.
  • Pick board: AI-Thinker ESP32-CAM.
  • Upload sketch, then remove the IO0 jumper and reset.

03MJPEG streaming

The bundled CameraWebServer example exposes /stream that pushes MJPEG over HTTP. Open the printed IP in any browser to see live video at up to 800×600 at ~15 FPS.

04On-device face detection

The same example includes a Face Detection toggle that uses Espressif's tiny CV library. It detects (and optionally recognizes) faces directly on the chip — no cloud needed for basic use cases.

05Sending frames to the cloud

For richer inference, capture a JPEG with esp_camera_fb_get(), then HTTP POST it to a vision API (Google Cloud Vision, AWS Rekognition, or a self-hosted YOLO server). Use deep sleep between captures for battery use.

06Real-world projects

  • Wildlife trail camera with PIR wake-up.
  • Doorbell that emails a snapshot.
  • Plant health monitor with daily upload.
  • License plate logger for a private gate.
Related part

Seminars & Reports