A Deep Learning Model for Video-Based Human Fall Detection Using CRNN and EfficientNet

People as they age, activities can become increasingly difficult, and they may require more assistance to carry out their routine tasks. This is because of physical, mental, and cognitive limitations that can affect their ability to perform daily activities. Social connections are critical to both mental and physical health. When seniors are lonely, they can experience real physical symptoms. Their blood pressure drops too much when they get up from sitting or lying down, they have balance and gait issues and they are prone to fainting and falling because of age-related muscle loss. It is generally considered as the sign of declining physical strength, balance issues, or other health-related concerns that require professional attention. As in some cases, they might require immediate actions to be taken due to potential for serious consequences. Continues monitoring, hospitalization might lead to long term disabilities and health care cost. This paper addresses this problem by introducing Deep Learning-Based Video-Based Human Fall Detection which plays a crucial part in promoting immediate attention and identification of falls, particularly among elderly individuals, by leveraging technology to monitor and respond quickly to such events. This serves as an effective solution to ensure immediate intervention in case of emergencies. Deep learning techniques are utilized to monitor an individual's physical appearance in real-time for fall detection. An alert is triggered if the person collapses or exhibits any significant deviation from their normal posture. The pre-processing phase begins with the application of a Gaussian filter to reduce noise, followed by feature extraction using the EfficientNet model for accurate analysis. Convolutional neural networks (CNN) and recurrent neural networks (RNN) are combined in the detection process to efficiently analyse temporal and spatial features from live video. Upon detecting a fall, an alert is promptly sent to the registered user's contact number via the Twilio API, ensuring timely intervention.