240MHz

LX7 DUAL-CORE

16MB

FLASH

8MB

OPI PSRAM

OV
3660

3MP CAMERA

BLE
5.0

+ WiFi

Native
USB

NO DRIVER

📷

3MP Resolution — Up to 2048×1536

Supports QXGA (2048×1536) for still image capture, UXGA (1600×1200) for high-quality stills, SXGA (1280×1024), XGA (1024×768), SVGA (800×600) and VGA (640×480) for smooth 30fps streaming.

Use VGA for streaming, UXGA for photo capture

🌞

Better Low-Light Performance

OV3660 has a larger pixel size compared to OV2640, capturing more light per pixel. This means less noise and more usable images in dim indoor environments — critical for indoor security cameras and baby monitors.

Pair with the onboard LED flash for complete darkness

🌟

Superior Colour Accuracy

OV3660 features an improved ISP (Image Signal Processor) with better auto white balance and colour reproduction. Face detection and recognition models achieve higher accuracy because faces are rendered with correct skin tones.

Better colour = better AI model accuracy

⏳

30fps at VGA / SVGA

Achieves smooth 30fps at VGA (640×480) and SVGA (800×600) resolutions, enabling fluid live video streaming with the ESP32-S3. The 8MB PSRAM provides enough buffer for multiple frames simultaneously.

Use double-buffering in PSRAM for smoothest stream

⚙️

Xtensa LX7 Dual-Core 240MHz

LX7 is 40% faster per clock than the LX6 in classic ESP32. One core handles WiFi/camera, the second handles AI inference simultaneously. No lag or frame drops during face detection.

LX7 vs LX6: 40% faster IPC, better power efficiency

🤖

AI Vector Instructions (PIE)

Espressif’s Processor Instruction Extensions (PIE) accelerate 8-bit vector operations used in neural networks. Face detection runs 2–5× faster than on classic ESP32, enabling real-time results at 10+ FPS.

Use ESP-WHO + ESP-DL frameworks to access PIE

💾

16MB Flash + 8MB OPI PSRAM

16MB Flash stores large AI models, web UI assets, and complex firmware. 8MB PSRAM via high-speed Octal SPI bus holds multiple camera frame buffers simultaneously — essential for smooth streaming.

Must set PSRAM to OPI PSRAM in Arduino IDE Tools

🔌

Native USB OTG on GPIO 19/20

Full-speed USB OTG built into the chip on GPIO 19 (D−) and GPIO 20 (D+). Acts as USB serial (CDC) for programming and Serial Monitor with no external chip. Can also act as USB mass storage or HID device.

Set USB CDC On Boot: Enabled for Serial.print() to work

📸 Frame Buffers

A single VGA frame (640×480 JPEG) needs ~50–100KB. Storing 2–4 frames simultaneously enables smooth streaming without dropped frames. PSRAM provides this buffer space that internal SRAM (512KB) simply cannot.

Result: smooth 30fps WiFi video stream

🤖 AI Model Storage

Face detection models (FaceNet, etc.) typically need 200–500KB of RAM at inference time. Without PSRAM these models simply will not fit. With 8MB PSRAM the ESP32-S3 can load and run multiple models simultaneously.

Result: real-time face detection at 10+ FPS

📷 High-Resolution Capture

UXGA (1600×1200) raw frame needs approximately 2MB. This fits only in PSRAM. Without it you are limited to VGA resolution for captures. With 8MB PSRAM you can capture full 3MP still images and save to SD card.

Result: 3MP photo capture to SD card

👤 Face Detection (ESP-WHO)

Detect human faces in the camera frame in real time using Espressif’s ESP-WHO framework. Draws bounding boxes around detected faces. Runs at 10+ FPS at QVGA (320×240) on ESP32-S3 PIE hardware. Use case: detect when someone approaches a door, trigger alerts, count people in a room.

Library: github.com/espressif/esp-who | Works in Arduino + ESP-IDF

🔐 Face Recognition (Enroll + Match)

Enroll up to 7 faces in flash memory. On detection, compare live face to enrolled faces and unlock a relay or display name on screen. Full attendance system or door access control with no cloud service, no monthly fee, and no internet connection required.

Use case: WiFi door lock that opens only for registered faces

🏛 Object Detection (TinyML / ESP-DL)

Run custom trained TensorFlow Lite models converted to ESP-DL format. Detect cats, cars, packages, or any custom object class. Deploy models trained in Google Colab directly to the board over USB. The 16MB Flash stores models up to ~3MB.

Tool: ESP-DL framework for quantised model deployment

📤 Motion Detection + Alert

Compare consecutive frames in PSRAM to detect pixel changes. When motion exceeds a threshold, send a Telegram message, trigger a relay, flash an LED, save a snapshot to SD card, or send an HTTP POST to your home automation server — all using the onboard WiFi.

Use case: security camera that sends Telegram photo on motion

📶 WiFi 4 (802.11 b/g/n 2.4GHz)

🔌 BLE 5.0

📼

Windows 10/11

No driver — appears as COM port instantly

🍎

macOS

No driver — /dev/cu.usbmodem port

🐧

Linux

No driver — /dev/ttyACM0 port

Point the OV3660 at your door, gate, or room. Connect to home WiFi. Access a live MJPEG stream from any browser on your network using the CameraWebServer example. Add motion detection to send a Telegram photo alert whenever something moves. Why this board: 8MB PSRAM enables buffer for smooth 30fps stream even when WiFi is active.

Project: door camera that sends Telegram photo on motion detection

Enroll employee or student faces. When a registered face is detected, log timestamp to SD card and display name on a connected OLED display. Trigger a relay to open a door or turnstile. Why this board: PIE AI vector instructions run FaceNet model 2–5× faster than classic ESP32, enabling real-time recognition without cloud API.

Project: offline face recognition door lock with SD card log

Use PIR sensor to wake from deep sleep on motion, capture 3MP still image with OV3660, save to SD card with timestamp, then return to deep sleep for long battery life. Solar panel charges a LiPo battery. Review images over WiFi connection. Why this board: 16MB Flash stores hundreds of captures. Deep sleep at ~10µA extends battery for weeks.

Project: solar-powered trail camera saving 3MP photos to SD card

Mount the camera on a rover chassis. Run TinyML object detection on the ESP32-S3. When a specific object (colour, shape, or class) is detected in frame, send steering commands over BLE 5.0 to the motor controller. Why this board: Dual-core LX7 allows camera processing on Core 0 and motor control on Core 1 simultaneously.

Project: TinyML obstacle-avoiding robot with colour object tracking

Place in a room with your baby or pet. Connect to home WiFi. View the OV3660 stream on phone browser from anywhere in the house. Add a sound sensor to auto-alert when crying is detected. LED flash for night view. Why this board: OV3660 better low-light image quality means clearer night image with minimal flash.

Project: WiFi baby monitor with sound alert and night LED

🤖 AI/ML Enthusiasts

Explore face detection, object classification, and TinyML on embedded hardware. The PIE vector instructions and 8MB PSRAM make real AI inference possible at 1549 BDT.

🏠 Home Security Builders

Build a DIY WiFi security camera, doorbell camera, or motion-triggered alert system at a fraction of commercial smart camera cost.

📚 Engineering Students

Final year EEE/CSE projects on computer vision, smart surveillance, IoT systems, and embedded AI. The ESP32-S3 is the most capable board in this price range for thesis projects.

🔧 Advanced Makers

Already built with Arduino and ESP32? The S3 CAM is your next step — native USB, AI acceleration, and 3MP camera for serious vision projects.

ARDUINO IDE 2.x → TOOLS MENU — ESP32-S3 CAM OV3660 EXACT SETTINGS

len, fb->width, fb->height);\n esp_camera_fb_return(fb);\n}\n\nvoid loop() {}’;var b=new Blob([c],{type:\”text/plain\”});var u=URL.createObjectURL(b);var a=document.createElement(\”a\”);a.href=u;a.download=\”s3cam_init.ino\”;document.body.appendChild(a);a.click();document.body.removeChild(a);URL.revokeObjectURL(u);})()” style=”background:#1a56db;color:#fff;border:none;border-radius:8px;padding:8px 18px;font-size:12px;font-weight:700;cursor:pointer;white-space:nowrap;”>⬇ Download .ino

// ESP32-S3 CAM OV3660 -- Camera Init + Capture -- Dream RC
#include "esp_camera.h"

void setup() {
  Serial.begin(115200);
  camera_config_t config;
  config.pixel_format = PIXFORMAT_JPEG;
  config.frame_size   = FRAMESIZE_VGA;   // 640x480
  config.jpeg_quality = 12;
  config.fb_count     = 2;               // double buffer PSRAM
  config.fb_location  = CAMERA_FB_IN_PSRAM;
  // ... set all pin definitions ...
  esp_camera_init(&config);

  camera_fb_t *fb = esp_camera_fb_get();
  Serial.printf("Captured %u bytes at %dx%d\n", fb->len, fb->width, fb->height);
  esp_camera_fb_return(fb);
}

Download the full .ino for complete pin definitions. Check your board schematic for exact GPIO assignments — they vary by manufacturer.

// WiFi Stream -- Dream RC Bangladesh
#include <WiFi.h>

const char* ssid = "YOUR_WIFI_NAME";
const char* pass = "YOUR_WIFI_PASSWORD";

void setup() {
  Serial.begin(115200);
  // init camera with PSRAM frame buffers (see Example 1)
  WiFi.begin(ssid, pass);
  while (WiFi.status() != WL_CONNECTED) { delay(500); }
  Serial.print("Stream at: http://");
  Serial.println(WiFi.localIP());
  // Start MJPEG HTTP server (see CameraWebServer example)
}

For complete code: File → Examples → ESP32 → Camera → CameraWebServer in Arduino IDE. Set CAMERA_MODEL_ESP32S3_EYE or match your board pinout.

\n\nstatic camera_fb_t *prevFrame = nullptr;\nconst int MOTION_THRESHOLD = 30; // pixel difference threshold\nint fileIndex = 0;\n\nbool motionDetected(camera_fb_t *curr) {\n if (!prevFrame) return false;\n int diff = 0;\n // Simple pixel difference (sample every 10th pixel for speed)\n for (int i = 0; i < curr->len && i < prevFrame->len; i += 10) {\n diff += abs((int)curr->buf[i] – (int)prevFrame->buf[i]);\n }\n return (diff / (curr->len / 10)) > MOTION_THRESHOLD;\n}\n\nvoid saveToSD(camera_fb_t *fb) {\n String path = \”/photo_\” + String(fileIndex++) + \”.jpg\”;\n File file = SD_MMC.open(path, FILE_WRITE);\n if (file) { file.write(fb->buf, fb->len); file.close();\n Serial.println(\”Saved: \” + path); }\n}\n\nvoid setup() {\n Serial.begin(115200);\n SD_MMC.begin(); // Init SD card\n // Init camera (see Example 1)\n}\n\nvoid loop() {\n camera_fb_t *fb = esp_camera_fb_get();\n if (!fb) return;\n if (motionDetected(fb)) {\n Serial.println(\”Motion detected! Saving…\”);\n saveToSD(fb);\n }\n if (prevFrame) esp_camera_fb_return(prevFrame);\n prevFrame = fb;\n delay(200);\n}’;var b=new Blob([c],{type:\”text/plain\”});var u=URL.createObjectURL(b);var a=document.createElement(\”a\”);a.href=u;a.download=\”s3cam_motion_sd.ino\”;document.body.appendChild(a);a.click();document.body.removeChild(a);URL.revokeObjectURL(u);})()” style=”background:#b45309;color:#fff;border:none;border-radius:8px;padding:8px 18px;font-size:12px;font-weight:700;cursor:pointer;white-space:nowrap;”>⬇ Download .ino

// Motion Detect + SD Save -- Dream RC Bangladesh
#include "SD_MMC.h"

void loop() {
  camera_fb_t *fb = esp_camera_fb_get();
  if (motionDetected(fb)) {
    // Save JPEG to SD card with timestamp
    File f = SD_MMC.open("/photo.jpg", FILE_WRITE);
    f.write(fb->buf, fb->len); f.close();
    Serial.println("Motion captured!");
  }
  esp_camera_fb_return(fb);
  delay(200);
}

\n#include \n#include \n#include \n\nconst char* ssid = \”YOUR_WIFI\”;\nconst char* password = \”YOUR_PASS\”;\nconst char* BOT_TOKEN = \”YOUR_BOT_TOKEN\”;\nconst char* CHAT_ID = \”YOUR_CHAT_ID\”;\n\nWiFiClientSecure client;\nUniversalTelegramBot bot(BOT_TOKEN, client);\n\nvoid sendPhoto() {\n camera_fb_t *fb = esp_camera_fb_get();\n if (!fb) return;\n bot.sendPhotoByBinary(CHAT_ID, \”image/jpeg\”, fb->len,\n [](uint8_t *buffer, size_t len, void *data) -> size_t {\n camera_fb_t *f = (camera_fb_t*)data;\n memcpy(buffer, f->buf, f->len);\n return f->len;\n }, nullptr, fb);\n esp_camera_fb_return(fb);\n Serial.println(\”Photo sent to Telegram!\”);\n}\n\nvoid setup() {\n Serial.begin(115200);\n // Init camera (see Example 1)\n WiFi.begin(ssid, password);\n while (WiFi.status() != WL_CONNECTED) delay(500);\n client.setInsecure(); // Skip SSL verification\n bot.sendMessage(CHAT_ID, \”ESP32-S3 CAM Online\”, \”\”);\n}\n\nvoid loop() {\n // Add motion detection logic here, then:\n sendPhoto();\n delay(10000); // Limit to one photo per 10 seconds\n}’;var b=new Blob([c],{type:\”text/plain\”});var u=URL.createObjectURL(b);var a=document.createElement(\”a\”);a.href=u;a.download=\”s3cam_telegram.ino\”;document.body.appendChild(a);a.click();document.body.removeChild(a);URL.revokeObjectURL(u);})()” style=”background:#7c3aed;color:#fff;border:none;border-radius:8px;padding:8px 18px;font-size:12px;font-weight:700;cursor:pointer;white-space:nowrap;”>⬇ Download .ino

// Telegram Motion Alert -- Dream RC Bangladesh
#include <UniversalTelegramBot.h>

void sendPhoto() {
  camera_fb_t *fb = esp_camera_fb_get();
  bot.sendPhotoByBinary(CHAT_ID, "image/jpeg",
    fb->len, /* binary upload callback */ nullptr, nullptr, fb);
  esp_camera_fb_return(fb);
  Serial.println("Photo sent to Telegram!");
}

Libraries needed: UniversalTelegramBot + ArduinoJson (install via Library Manager). Create a Telegram bot via @BotFather to get your token.

🔌 MFRC-522 RFID Module

Add RFID card scanning to your face recognition access system for two-factor security

View →

⚡ Relay Module

Unlock a door, trigger a siren, or cut power when face is recognised or motion detected

View →

🔴 IR Sensor Module

Trigger camera capture when someone breaks the IR beam — great for trail cameras

View →

📺 OLED 0.96” Display

Display face recognition results, IP address, and system status on a compact I2C screen

View →

🌦️ DHT22 Temperature Sensor

Add temperature and humidity overlay to camera stream for environment monitoring

View →

🚏 HC-SR04 Ultrasonic Sensor

Detect approach distance — trigger camera capture when someone gets within range

View →

1 × ESP32-S3 N16R8 CAM Development Board with OV3660 Camera