Distance and Angle Insensitive Radar-Based Multi-Human Posture Recognition Using Deep Learning.

Area of Science:

• Utilizes radar signal processing and deep learning for human-computer interaction.
• Focuses on computer vision and pattern recognition in healthcare technology.

Background:

• Human posture recognition is crucial for preventive healthcare and monitoring.
• Existing radar-based methods face challenges with static human subjects and multiple individuals.

Purpose of the Study:

• To develop a novel framework for recognizing the postures of two humans in close proximity using FMCW radar point clouds.
• To enable privacy-preserving healthcare sensing for monitoring individuals, such as elderly couples.

Main Methods:

• Extracts range, velocity, and angle information from FMCW radar to create Cartesian point clouds.
• Employs unsupervised clustering to segregate individual human point clouds.
• Applies a deep learning model (DenseNet) for posture classification of individual subjects.

Main Results:

• Achieved an average accuracy of 96% in classifying ten posture combinations for two non-overlapping humans (standing, sitting on chair, sitting on floor, lying down).
• Introduced an approach using centroid information to detect and classify overlapping human participants.
• Yielded an accuracy above 96% for five overlapping human posture combinations.

Conclusions:

• The proposed FMCW radar framework effectively recognizes human postures for multiple individuals, including overlapping scenarios.
• Demonstrates potential as a privacy-preserving sensing platform for remote healthcare monitoring.
• Advances the application of radar technology in the detective and preventive healthcare industry.