Implementing a Hand Gesture Recognition System Based on Range-Doppler Map.

Area of Science:

• Human-Computer Interaction
• Radar Signal Processing
• Machine Learning

Background:

• Traditional vision-based hand gesture recognition systems for human-machine interfaces often raise privacy concerns.
• Non-vision-based techniques are increasingly researched to overcome the limitations of camera-dependent methods.
• Contactless control is a growing area of interest for intuitive device interaction.

Purpose of the Study:

• To propose a novel dynamic hand gesture recognition system utilizing 60 GHz FMCW radar.
• To develop a privacy-preserving, non-vision-based solution for contactless device control.
• To achieve high accuracy in classifying hand gestures using radar signatures.

Main Methods:

• Utilized 60 GHz FMCW radar to capture hand gesture signals.
• Transformed raw radar data into range, velocity, and angle domains for feature extraction.
• Developed and employed an end-to-end deep learning model, combining neural networks and long short-term memory (LSTM), for gesture classification.
• Used a camera solely for supporting the labeling of training data.

Main Results:

• The proposed system successfully transforms radar signals into human-understandable domains.
• The deep learning model effectively extracts features and classifies hand gestures.
• Achieved a high classification accuracy of 98% for dynamic hand gestures.
• Demonstrated the system's adaptability to different scenarios through customizable signatures.

Conclusions:

• The 60 GHz FMCW radar-based system offers a viable and accurate solution for dynamic hand gesture recognition.
• This non-vision approach effectively addresses privacy concerns associated with traditional methods.
• The developed deep learning model shows significant potential for robust contactless device control applications.