HIT HAR: Human Image Threshing Machine for Human Activity Recognition Using Deep Learning Models.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Human Activity Recognition (HAR) is crucial for healthcare systems, aiding in health monitoring and abnormal activity prediction.
• Conventional HAR systems using wearable sensors (IMU, stretch sensors) struggle with complex activities due to sensor errors, leading to misclassification.
• Existing radiofrequency and vision-based HAR systems also face real-time classification challenges.

Purpose of the Study:

• To address the limitations of current HAR systems by proposing a novel Human Image Threshing (HIT) machine.
• To leverage smartphone camera image datasets for accurate human activity recognition.
• To enhance the reliability and accuracy of HAR for improved healthcare applications.

Main Methods:

• The proposed HIT machine utilizes a Mask Region-based Convolutional Neural Network (R-CNN) for human body detection.
• A Facial Image Threshing (FIT) machine is employed for image cropping and resizing.
• A deep learning model, specifically ResNet architecture, is used for the final activity classification.

Main Results:

• The HIT machine-based HAR system demonstrated high effectiveness through extensive experiments.
• The system achieved a remarkable accuracy of 98.53% when utilizing the ResNet architecture.
• This accuracy signifies a significant improvement over conventional HAR methods, especially for complex activities.

Conclusions:

• The proposed HIT machine offers a robust and accurate solution for human activity recognition using readily available smartphone cameras.
• This approach effectively overcomes the misclassification issues associated with sensor-based HAR systems.
• The HIT machine holds significant potential for advancing remote health monitoring and personalized healthcare.