A real-time fall detection system based on the QuecPython embedded platform, using the MPU6500 IMU sensor and random forest classifier.

This TinyML system applies machine learning directly on embedded hardware to detect movements in real time. It adopts a small sensor named MPU6500 to measure motion and rotation, and is currently optimized for X-axis and Y-axis movements.

• Real-time Detection: Response time for motion detection is 150-300ms
• Memory Efficiency: Automatic buffer management with overflow prevention
• False Alarm Prevention: 3 consecutive consistent results are required before detection
• Timer-based Processing: Non-blocking hardware timer architecture
• Motion Segregation: Each motion is analyzed independently to avoid mutual interference

tinyml-Fall/
├── src
	├── _main.py              # Main application based on the timer system
	├── mpu6500.py            # MPU6500 sensor driver with m/s² scaling
	├── random_forest.py      # Pre-trained random forest model
	├── tinyml.py             # TinyML pipeline with anti-shake mechanism
├── LICENSE
├── README.md
└── README_zh.md

• Platform: Quectel embedded module running MicroPython
• Sensor: MPU6500 6-axis IMU (3-axis accelerometer + 3-axis gyroscope)
• Model: Random forest classifier (3 categories: 0 = Static, 1 = Walking, 2 = Fall type)
• Sampling: 50Hz sensor reading, 20Hz inference
• Detection: 3 consecutive consistent results within a 450ms window
• Data Format: Accelerometer (m/s²), Gyroscope (deg/s)

• X/Y-axis Motion - Reliable operation and stable detection
• Memory Management - Buffer contamination issue has been resolved
• Real-time Processing - Timer-based architecture has been implemented
• Anti-shake System - False alarm prevention function works properly
• Circular Motion - Currently under development

Before you start, please ensure you have the following prerequisites:

• Hardware:
  • One QuecPython development board
  • A computer (Windows 7, Windows 10, or Windows 11)
  • MPU6500 sensor
• Software:
  • Debugging tool QPYcom
  • QuecPython firmware (The EG912U firmware can be found in the fw directory of the repository)
  • Python text editor (e.g., VSCode, PyCharm)

1. Clone the repository:

git clone https://github.com/aaronchenzhihe/Tinyml-Fall.git

2.Flash the firmware: Flash the firmware to the development board according to the instructions.

1. Hardware Connection: Correctly connect the MPU6500 sensor to the I2C interface of the development board.
2. Connect to the host computer via Type-C cable.
3. Download the code to the device:
   • Launch the QPYcom debugging tool.
   • Connect the data cable to the computer.
   • Press the PWRKEY button on the development board to power it on.
   • Import all files in the src folder into the module's file system while maintaining the directory structure, following the instructions.
4. Run the code
   • Select the File tab.
   • Select the _main.py script.
   • Right-click and select Run to execute the script.

The system automatically detects motions in real time. When a motion is recognized, it outputs the classification result (0, 1, or 2) and clears all buffers to prevent interference from previous motions.

• Detection Latency: Theoretical minimum of 150ms, actual 200-300ms
• Memory Usage: Total < 50KB
• Accuracy: Optimized for X/Y-axis movements