Human Activity Recognition using 1D-CNN and Stacked LSTM
Sheetal Waghchaware, Radhika D. Joshi · 2022
Computer vision is an engrossing area of Artificial Intelligence that is growing continuously due to huge amounts of data uploaded online daily. Human Activity Recognition (HAR) is an extensively considered problem in Computer vision. Although a lot of research has been done in the field of activity recognition, still it is a challenging job for researchers to enhance the performance of the network as there are many limitations such as variation in illumination, complex nature of human activities, background cluttering, occlusion, many activities at the same time. HAR is a process of identifying human activities in an environment obtained from data through various modalities like camera, inertial measurement unit (IMU), body-worn, physiological, positional, and environmental sensors. A hybrid model of 1D-CNN and stacked LSTM is proposed for HAR. The proposed method is tested on the WISDM dataset and it gives superior performance compared to the state- of-art-techniques. It has achieved an accuracy of 97%. The analysis is done based on hyperparameters, methodology, and performance metrics to validate the results.