Abstract:
With the development of various sensors in smartphones or wearable devices, human activity recognition (HAR) has been widely researched and has many applications in healthcare, smart cities, etc. Sensor-based HAR uses many hand-crafted feature engineering or deep neural network methods. These methods recognize activities offline, requiring large storage space to collect all data before training. The offline model training also requires retraining to recognize new activities, which takes time and space. This paper proposes HarMI, a multi-modality incremental learning model with continuous learning. The HarMI model can quickly start training and learn new activities without storing previous training data. First, we align heterogeneous sensor data with different frequencies using attention mechanism. HarMI uses elastic weight consolidation and multi-modality canonical correlation analysis to overcome catastrophic forgetting in incremental learning. HarMI outperforms several state-of-the-arts in extensive experiments on two public datasets.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
