Advanced artificial intelligence with a two-tier optimization framework based on smart human activity recognition on assistive technologies for disabled people
Manal Abdullah Alohali, Mohammed Yahya Alzahrani, Asmaa Mansour Alghamdi, Ishfaq Yaseen · Journal of the Chinese Institute of Engineers · 2025
Human activity recognition (HAR) plays a crucial role in various applications such as healthcare, sports, surveillance, and rehabilitation. It involves analyzing human motions using data from wearable or mobile sensors and vision-based systems. Recently, deep learning (DL) approaches have demonstrated outstanding performance in interpreting HAR data. This study proposes the Two-Tier Optimized Artificial Intelligence Framework for Smart Human Activity Recognition in Assistive Technologies (TTOAI-SHARAT) to enhance real-time activity recognition for individuals with disabilities. The TTOAI-SHARAT framework begins with Z-score normalization to standardize the input data, ensuring consistency across diverse activity patterns. It then employs the sparrow search algorithm (SSA) for efficient feature selection, which minimizes computational load by identifying the most significant attributes. For classification, the model integrates a convolutional sparse autoencoder (CSAE), which effectively captures key features from sensor data. To further refine the model’s performance, a multi-spiral whale optimization algorithm (MSWOA) is used for hyperparameter tuning, optimizing the CSAE’s analytical accuracy. The framework was evaluated using a benchmark dataset, and the results demonstrated superior performance across various metrics, confirming its robustness and reliability. TTOAI-SHARAT provides a promising solution for smart assistive technologies, offering accurate and efficient recognition of human activities in real-time scenarios.