Skip to main content

Human Face Detection

Overview​

This example demonstrates human face detection based on deep learning and provides a lightweight, extensible framework for face-related applications.

The project is developed using theESP-DL library,which offers deep learning capabilities within the ESP-WHO framework. With support for different hardware modules, it can be used in various AIoT scenarios.

Development Preparation​

Hardware​

  • NE100-MB01 Development Board

Software​

1. Firmware：​

Download pre-compiled firmware for use：

2. Environment Requirements​

To perform secondary development, make sure the following：

  • IDE Software：Visual Studio Code（v1.99.2+）
  • Development Framework：ESP-IDF Extension（v5.1.6）
  • Example Repository：camthink-ai/esp-who

Important：Please follow the Development Environment Setup Guide to complete the initial configuration.

Functionality Verification​

Using Precompiled Firmware​

1. Hardware Connection​

Connect the development board to PC using a Type-C cable.

NE101 interface

2. Firmware flashing​

Refer to the detailed flashing guide：

3. UART Output Analysis​

The device outputs face detection results via the serial port (baud rate: 115200)：

Example Code:

WhoDetect: 0, bbox: [0.826712, 47, 40, 169, 195], left_eye: [80, 105], left_mouth: [90, 151], nose: [101, 125], right_eye: [123, 99], right_mouth: [126, 146]

Field Descriptions：

  • bbox：Face bounding box information
    • Format：[confidence, top-left x, top-left y, width, height]
  • left_eye：Coordinates of the left eye(x,y)
  • right_eye： Coordinates of the right eye (x,y)
  • nose：Coordinates of the nose(x,y)
  • left_mouth：Coordinates of the left mouth corner(x,y)
  • right_mouth：Coordinates of the right mouth corner(x,y)

4. Performance Test​

  1. Point the camera at a real human face or a face image.
  2. Observe the serial output for facial landmark data.
  3. Move the camera around to test dynamic face detection performance.

Static Illustration of Face Detection

Real-Time Face Detection Example

Source Code Development​

1. Clone the repository using the following command​

git clone https://github.com/camthink-ai/esp-who

2. Project setup​

  1. Open the directory in VS Code：examples/human_face_detect

    Project Root Directory

  2. Select esp32-s3 as the target chip：

    芯片选择界面

3. Compile and deploy the source code​

  1. Select the build button to compile the source code：

    Esp32s3 selection

  2. Select the flash button to upload the generated firmware to the device：

    Showing build output

4. Performance Test​

Same with using precompiled firmware

References​

  1. ESP-DL Model Quantization Documentation：

  2. Official Example Repository：