Human activity recognition (HAR) plays a crucial role in people’s daily life for its wide range of applications
Two main types of HAR:
- Video-based HAR: analyzes videos or images containing human motions from the camera
- Sensor-based HAR: motion from sensors – accelerometer, gyroscope, Bluetooth, sound sensors, etc.
HAR using wearable devices has been actively investigated for a wide range of applications:
- Healthcare: fall detection systems, elderly monitoring, and disease prevention
- Sports training: energy expenditure, skill assessment
- Smart assistive technologies, i.e. smart homes: aid people with cognitive and physical limitations, etc.
-
Focus on Sensor-based HAR: using accelerometer data to classify 6 activities
-
Apply different types of Deep Learning technique to discover which method performs the best in term of: Generalization, Accuracy, f1-score, precision, recall, time given minimal data- preprocessing & transformation
- Wireless Sensor Data Mining Lab (Fordham University) https://www.cis.fordham.edu/wisdm/dataset.php
Reference: https://www.cis.fordham.edu/wisdm/includes/files/sensorKDD-2010.pdf
- DNN (MLP)
- LSTM + Dense
- LSTM stacked 3 layers
- CNN-LSTM
- ConvLSTM