An Edge Computing Based Fall Detection System for Elderly Persons

摘要

Image processing based fall detection has been widely studied. However, due to the different customers and rooms, the system may not work well with the prepared training data. Therefore, a fall detection system which can automatically adapt the environment is important. It is a good solution that system selects the most suitable model to detect the fall and uses the daily life images of the users to re-train the model while mis-detection occurs. This work studies how to select the most suitable model to detect the fall. Moreover, if a mis-detection occurs, the mis-detected frames are uploaded to cloud server and the model is retrained and sent to the edge node to detect the fall. Deep learning methods are employed to test the training data, and GoogLeNet gives the best performance.