Fall Detection for the Elderly Based on Online Transfer Learning

Fall detection is of great significance to the elderly. The wearable fall detection device can collect real-time data to identify falling events, thereby helping to protect the elderly from suffering further injuries. The limited sensor data from older adults in the SisFall dataset is insufficient for training a fall detection classifier that is specifically tailored to older adults. This paper proposes a fall detection method based on online transfer learning. The method uses the weighted online sequential extreme learning machine with a forgetting factor as an online classifier, which can effectively update the model in real time based on continuously collected data, thereby improving the accuracy of fall detection among elder adults. By dynamically updating the combination weights of offline classifiers and online classifiers to transfer source domain knowledge to the target domain, the proposed method improves classification accuracy in the target domain. Moreover, we incorporate concept drift detection to adapt to changes in the data distribution over time. Experimental results show that the improved algorithm has a higher online accuracy.