-
📱 the mobile application for Android Systems (v. 4.0 and higher)
-
👨🔬 using Convuctional Neural Networks to classificate motion activities
-
📚 using Smartphone-Based Recognition of Human Activities and Postural Transitions Data Set for the training and testing CNN model
-
📳 real-time inertial signals registration (based on accelerometer and gyroscope)
- Python v. 3.10.1
- TensorFlow v. 2.9.1 n Keras
- Jupyter Notebook
- Java v. 8
- Android Studio v. 2020.3.1
This project has three main steps:
raw signals processing, noises filtering: 3rd order Butterworth filter, separating the gravity component: high-pass filter, windowing signals, normalisation, division of the data into training and testing components, matching the shape to the CNN model, signals visualisation
- Example of the Data Frame with values after the signals processing:
- Histogram with the obtained data samples:
designing a CNN model, training and testing process, fitting and evaluating, parameters visualisation
- CNN model:
- Quality of the CNN model:
🔬 MODEL ACCURACY: 95,03%
the mobile application based on CNN model to clasificate 6 motion activities in real-time:
1. WALKING 2. WALKING UPSTAIRS 3. WALKING DOWNSTAIRS 4. SITTING 5. STANDING 6. LAYING
- Application Interface:
🙋 How to place the phone on the body:
📱 Phone axes:
The application has been tested on a group of people aged 25, 27 and 45. The results of the above program were satisfactory. The application is very good at classifying activities with indexes 1, 3, 4 and 6, while it makes occasional errors when classifying activities with indexes 2 and 5. This model is an excellent basis for further research into creating a useful application for motion recognition.
