Co-Saliency-Enhanced Deep Recurrent Convolutional Networks for Human Fall Detection in E-Healthcare

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Fall detection is crucial for e-healthcare and assisted-living applications.
• Conventional methods rely on hand-crafted features, which can be limiting.
• Deep learning offers potential for more robust fall detection from video data.

Purpose of the Study:

• To propose a novel fall detection scheme using a co-saliency-enhanced recurrent convolutional network (RCN).
• To improve the accuracy and reliability of fall detection in video streams.
• To enhance the performance of deep learning models by integrating co-saliency mechanisms.

Main Methods:

• Developed a Recurrent Convolutional Network (RCN) architecture combining Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM).
• Integrated a co-saliency-based method to enhance salient human activity regions within video frames.
• Conducted extensive empirical tests on an open dataset with multi-camera videos.

Main Results:

• The proposed RCN architecture achieved a high test accuracy of 98.96% for fall detection.
• Co-saliency enhancement significantly improved the deep learning model's performance.
• The scheme demonstrated superior performance compared to two existing fall detection methods.

Conclusions:

• The co-saliency-enhanced RCN is an effective deep learning approach for video-based fall detection.
• This method shows significant promise for applications in e-healthcare and assisted-living.
• The proposed architecture offers a robust and accurate solution for automated fall monitoring.