Fall Event Detection using Vision Transformer

Abstract

Privacy-preserving radar-based fall detection for older adults is becoming essential as falls in adults above 65 years of age may result in death or a permanent physical disability. In this paper, a novel deep learning-based fall event detection technique using Vision Transformers with Shifted Patch Tokenization and Locality Self Attention is proposed. The proposed approach is evaluated using publicly available dataset. Preliminary evaluation shows improved performance over transfer learning models and standard Vision Transformer.