Human Activity Detection using Machine Learning Cascade Architecure by using TensorFlow lite for microcontrolers, u-blox cellular product and libraries.

Description

You can use detect human activity using TensorFlow lite for microcontroller. By using u-blox cellular products & libraries, you can send data from remote devices to your server where a bigger model is deployed, hence human activity can be detected in better way in this ML Cascade Architecture. This project is being added to help u-blox customers for rapid prototyping and evaluating u-blox products. This repository is tested using C030-R412M board and Sparkfun's Asset tracker Micromod setup using STM32F405 microcontroller.

This repo uses Git submodules: make sure that once it has been cloned you do something like:

git submodule update --init --recursive

Note-1 : You may need to add servers details or set your own URI, if you want to transfer data to cloud.

Note-2 : If you want to add your own accelerometer then remove +20 test data of accelerometer and feed accelerometer data to Model at 20 Hz to get inference.

Note-3 : By default only classified accelerometer test data are being feed to model in this firmware.

Build using build configuration through CMake.

Use following two commands to build the firmware.

cmake -S . -B build/debug --warn-uninitialized -DCMAKE_BUILD_TYPE=DEBUG -DCMAKE_TOOLCHAIN_FILE=toolchain-STM32F405.cmake -GNinja

cmake --build build/debug

Flash the hex file & start executing the firmware. It is a ready to build and execute example.

You should build and run this example firmware in 10 minutes if you are not doing something wrong!

Hardware requirements

HW type	Name
Carrier Board	SparkFun MicroMod Asset Tracker
Application board	MicroMod_STM32_Processor
Cellular	SARA-R5 Module
Microcontroller	STM32F405RGT6

Build requirements

Toolchain or Build Software	Version
CMake	3.25.1
Ninja	1.11.1
GNU Tools for ARM Embedded Processors	6-2017-q2-update

Memory requirements of Firmware

Memory region	Used Size	Region Size	%age Used
CCMRAM	36 B	64 KB	0.05%
RAM	80320 B	128 KB	61.28%
FLASH	497560 B	1 MB	47.45%