Abnormal Behavior Detection of the Solitary Elderly in Rural Areas Based on Lightweight MotionBERT
Hui Xie, Qihe Fang, Zhiwang Xu · IEEE Access · 2025
The aging of the population is reshaping human societal structures at an unprecedented pace and scale that has become one of the most critical challenges of the 21st century. Rural areas may confront the more severe demographic structure issue. Therefore, identifying and preventing abnormal behavior of the solitary elderly in rural areas is paramount to enhancing their safety and promoting digital rural construction. Deep learning can effectively deal with large-scale data and learn complex nonlinear relationships, which can monitor and early-warn the daily activities, postural patterns, and health risks of the solitary elderly in rural areas. However, the widespread deployment faces significant challenges due to the excessive computational demands of traditional deep learning models. To address this issue, this paper proposed a human behavior recognition method based on lightweight MotionBERT architecture, which can deploy embedded devices to detect the behavior of solitary elderly. The spatiotemporal attention modules, redundant multi-head attention, and cross-entropy loss of MotionBERT architecture are optimized to reduce computational complexity and keep the model’s performance.