Abstract Falling poses significant risks, especially for the geriatric population. In this study, the authors introduce an innovative approach to privacy‐preserving fall detection using computer vision. The authors’ technique leverages a deep neural network (DNN) to accurately identify falling events in input images, while simultaneously prioritizing privacy through the implementation of an optical element. The experimental results establish that the authors’ proposed method outperforms alternative hardware and software‐based privacy‐preserving approaches in terms of encryption level and accuracy. These results are derived from an extensive dataset encompassing diverse falling scenarios.