Human activity recognition using CNN and LSTM for inertial sensors activity data
Abstract
Human Activity Recognition is used in diverse applications due to its vast domain of data acquisition sources. In today's digital world devices like closed circuit television (CCTV), devices built-in inertial sensor, Wi-fi are the prominent sources for activity data gathering. Applications of HAR includes healthcare monitoring, surveillance, smart home, and many more make HAR models a crucial research domain. But there are challenges due to the vast amount of varying data captured from varying sources. The evolution of Artificial Intelligence framework with traditional and advanced deep algorithms, created opportunities for researchers to build reliable and efficient HAR models. ML algorithms require lots of effort and time as they involve hand-crafted feature extraction. Deep Learning (DL) models are more suitable as it learns more meaning features during training. In the proposed study, we built two DL models 1D-CNN and CNN-LSTM using smartphone inertial sensor data. In this study we have achieved the 95.13% and 93.56% accuracy for heterogeneity HAR dataset. In proposed models we evaluated the model's performance using various performance evaluation metrics and used K-fold validation.
- Publication:
-
American Institute of Physics Conference Series
- Pub Date:
- March 2024
- DOI:
- 10.1063/5.0198752
- Bibcode:
- 2024AIPC.3072b0019G
- Keywords:
-
- TRACK-1 - ARTIFICIAL INTELLIGENCE (AI)