Accuracy Improvement Using Conditional GAN in FMCW Radar-Based Fall Detection

As the number of elderly individuals living alone increases, these elderly individuals face significant challenges in responding quickly to falls. To address this issue, research on human activity recognition for fall detection is crucial. In this study, we utilize frequency-modulated continuous wave radar combined with deep learning to enhance fall detection. However, considering that deep learning models depend on large datasets, the lack of collected fall data poses a major obstacle. To overcome this, we propose a method that integrates a conditional generative adversarial network and transfer learning. This approach resulted in improved overall performance, with a notable increase in fall detection accuracy.