Driving Activity Recognition Using UWB Radar and Deep Neural Networks.

Area of Science:

• Automotive Engineering
• Signal Processing
• Machine Learning

Background:

• In-car activity monitoring is crucial for automotive safety functions.
• Current methods predominantly rely on vision systems, which raise privacy concerns.
• Radar technology presents a low-cost, privacy-preserving alternative for activity monitoring.

Purpose of the Study:

• To introduce a novel approach for in-car activity monitoring using ultra-wideband (UWB) radar.
• To classify driving activities by leveraging the Doppler signal processed by deep neural networks.
• To enhance generalization to unseen individuals and provide a benchmark dataset for future research.

Main Methods:

• Utilizing the Doppler signal from an ultra-wideband (UWB) radar as input.
• Employing deep neural networks for the classification of driving activities.
• Developing and releasing a new radar driving activity dataset (RaDA) for public use.

Main Results:

• Demonstrated a novel method for in-car activity classification using UWB radar Doppler signals.
• Achieved classification of driving activities through deep neural network analysis.
• Established a new dataset (RaDA) to facilitate research and benchmarking in radar-based activity monitoring.

Conclusions:

• Radar-based systems offer a viable, privacy-preserving solution for in-car activity monitoring.
• Deep neural networks effectively classify driving activities using UWB radar Doppler signals.
• The RaDA dataset will spur further development and comparison of radar-based automotive safety technologies.