LapseNet: A Hybrid CNN-LSTM Approach for Accurate and Efficient Vision-Based Fall Detection

Falls are a major cause of injury and death among the elderly population, particularly in unsupervised settings where victims often remain unattended for extended periods of time. Such incidents can lead to long-term physical and mental disturbances such as fractures, skin burns, blood loss, and trauma. A reliable and effective fall detection system can ensure that support is provided immediately, improving chances of recovery for victims. A diverse range of fall detection methods have been studied and tested, but most have high false positive rates and limited robustness in real-world scenarios. In this study, we present LapseNet, a hybrid convolutional neural network with long short-term memory to detect falls in indoor settings. We utilized data from four publicly available sources, with a total of 250 videos for training and testing the model, which distinguishes between a) falls and b) activities of daily living. LapseNet achieved a training accuracy of 99.43% and a promising testing and validation accuracy of 100%. These results demonstrate the potential to significantly improve elderly care and safety by enabling timely interventions and reducing the risk of long-term complications from falls.