TL-FALL: Contactless Indoor Fall Detection Using Transfer Learning from a Pretrained Model

A new radar-based fall detection method is proposed using the recent advances in deep neural networks. An ultrawideband radar is used to monitor human daily activities and identify the occurrence of falls. A transfer learning approach is employed based on a pre-trained model on ImageNet dataset to realize a robust feature extraction from radar data. The architecture and depth of the model are fine-tuned to radar time-frequency representations. From the results, it is observed that the proposed transfer learning based method can achieve a detection accuracy for fall incidents higher that those of the other methods.