ADL

uTensor enables motion recognition on microcontrollers. The model is trained with a modified Activity of Daily dataset recognizing 5 classes:

• Walking
• Climbing
• Activities
• Descending
• Resting

The project is also a reference implementation of sequential data processing with Mbed and uTensor.

For sensor setup, please refer to Train/HMP_Dataset/MANUAL.txt. The grove sensor is place flat on the back of user's right hand, with the connector socket oriented furthest away from the wrist.

Hardware requirement:

• Mbed F413ZH board
• Grove Sheild
• Grove 3D digital accelerometer

Build Instruction

• Recommend cloud9 environment
• Run:

$ mbed import https://github.com/uTensor/ADL_demo
$ cd ADL_demo
$ mbed compile -m DISCO_F413ZH -t GCC_ARM --profile=uTensor/build_profile/release.json

• Ensure the Grove sensor is connected
• Locate the binary path from the terminal output, and flash it onto the board

Training

For Training Instruction, please see Train/README.md