Human Activity Recognition with a Time Distributed Deep Neural Network

Human activity recognition (HAR) is necessary in numerous domains, including medicine, sports, and security. This research offers a method to improve HAR performance by using a temporally distributed integration of convolutional neural networks (CNN) and long short-term memory (LSTM). The proposed model combines the advantages of CNN and LSTM networks to obtain temporal and spatial details from sensor data. The model efficiently learns and captures the sequential dependencies in the data by scattering the LSTM layers over time. The proposed method outperforms baseline CNN, LSTM, and existing models, as shown by experimental results on benchmark datasets UCI-Sensor and Opportunity-Sensor dataset and achieved an accuracy of 97% and 96%, respectively. The results open up new paths for real-time applications and research development by demonstrating the promise of the temporally distributed CNN-LSTM model for improving the robustness and accuracy of human activity recognition from sensor data.