Project Hub /Artificial Intelligence & Edge ML/Object Detection & Tracking/Edge AI Face Counting Using MaixCam and YOLOv8 for Real-Time Audience Analytics

Edge AI Face Counting Using MaixCam and YOLOv8 for Real-Time Audience Analytics

Intermediate

MAIXDUINO

Project Specific

3 – 6 Hours

Partial Kit

Edge AI Face Counting Using MaixCam and YOLOv8 for Real-Time Audience Analytics

An edge AI system built using the MaixCam platform and YOLOv8 for real-time face detection and audience counting in open spaces.

Originally published on ElectronicsForU

View original tutorial

Parts You'll Need

This project is sourced from another site, not written by us — this list is our best match of their components against what we stock, not a verified bill of materials. Always check the original blog (linked above) for the actual parts list before buying.

ESP32 CAM WiFi Module Bluetooth with OV2640 Camera Rs. 810.00
400 Points Mini Solderless Prototype PCB Breadboard High Quality Rs. 89.00
MB102 Adjustable Breadboard Power Supply Module (3.3V & 5V) for Arduino & Raspberry Pi Rs. 139.00
Jumper Wire 10 Pin Male To Male 20cm Rs. 45.00
Jumper Wire 10 Pin Male To Female 20cm Rs. 45.00

Selected: Rs. 0.00

We don't carry yet: SD / MicroSD Card
Wi-Fi)
file retrieval

Browse all components

How It Works

The Edge AI Face Counting system operates entirely on the MaixCam platform, leveraging its onboard neural processing unit (NPU) and 5M camera to perform real-time computer vision tasks without relying on external cloud servers. The system captures live video frames through the built-in camera and passes them to a specialized YOLOv8 face detection model (yolov8n_face.mud) optimized for edge deployment. The NPU accelerates the heavy matrix operations required by the convolutional neural network, allowing the device to efficiently detect human faces in open-space environments. As the model identifies and bounds faces in each video frame, a Python script running via MaixPy counts the detected bounding boxes in real time. The resulting analytics and visual feedback can be monitored directly on the MaixCam's integrated TFT touch screen. Additionally, developers can interface with the board via a laptop using Wi-Fi and SSH to retrieve captured image files, inspect logs, or adjust the underlying Python script parameters to refine the audience counting logic.

Why Build This

Deploy optimized computer vision models directly to resource-constrained edge hardware without cloud dependency.

Implement real-time object detection pipelines using modern neural network architectures like YOLOv8.

Develop practical audience analytics and crowd monitoring systems suitable for live events and public spaces.

Real-World Application

Live events, speeches, open-space crowd monitoring, and audience engagement analytics.

Skills You'll Learn

Edge AI deployment

Neural Processing Unit (NPU) acceleration

Object detection implementation

Python programming for embedded systems

SSH remote management

Computer vision data processing

Safety Precautions

No specific precautions noted for this build — always follow general electronics safety practices.

Technology Tags

EDGE AI

YOLOV8

PYTHON

FACE DETECTION

MACHINE LEARNING

RISC-V

EMBEDDED LINUX

Frequently Asked Questions

Do I need an internet connection for the face detection to work?

No, the entire YOLOv8 inference and face counting process runs locally on the MaixCam's built-in NPU, requiring network connectivity only if you choose to use SSH or Wi-Fi for file retrieval.

What is the purpose of the MicroSD card in this setup?

The MicroSD card is used to store system files, the Python scripts, and the YOLOv8 model weights (yolov8n_face.mud) that the MaixCam loads during operation.

Can I modify the counting script to detect objects other than faces?

Yes, though you would need to replace the face model with a different YOLOv8 model trained on a custom dataset and update your Python script to handle the corresponding class labels.

Gallery

This page is part of Kitkraft's Project Hub — a curated collection of electronics & robotics projects gathered from the maker community to help students and hobbyists discover and learn to build. Kitkraft does not claim authorship of the original project, and makes no representation as to the accuracy, completeness or safety of third-party source material. The summary and "How It Works" explanation above are AI-generated, rewritten with reference to the original tutorial — not a reproduction of it. Full credit and a direct link to the original source are provided above. If you're the original creator and have a concern about how your work is represented here, please contact us and we'll address it promptly.

Kitkraft Project Hub — curated from the maker community, credited at the source.

PRJ-01261