A New Method for Traffic Participant Recognition Using Doppler Radar Signature and Convolutional Neural Networks

Area of Science:

• Automotive Safety
• Artificial Intelligence
• Signal Processing

Background:

• Road accidents involving pedestrians and cyclists are increasing.
• Current vehicle safety systems require enhancement, particularly for autonomous vehicles.
• Advanced sensor technology and AI are crucial for improving situational awareness.

Purpose of the Study:

• To develop and evaluate a novel method for recognizing and distinguishing multiple objects (pedestrians, cyclists, cars) using radar signatures.
• To enhance road safety by improving the detection capabilities of vehicles, especially autonomous ones.
• To introduce a specialized convolutional neural network (CNN) architecture for radar data analysis.

Main Methods:

• Utilized frequency-modulated continuous-wave (FMCW) radar for object detection.
• Developed algorithms for time-frequency domain signal analysis and radar imaging.
• Created a custom database of radar signatures for pedestrian, cyclist, and car models in a Matlab environment.
• Employed a specialized deep neural network, specifically a convolutional neural network (CNN), for object recognition.

Main Results:

• Successfully demonstrated a method for recognizing and distinguishing groups of objects based on radar signatures.
• The proposed CNN architecture showed positive results in simulations and tests.
• The approach enables discrimination between multiple objects within a single radar signature.

Conclusions:

• The developed system shows significant potential for enhancing vehicle safety, particularly for autonomous driving.
• The innovative approach to generating custom input databases and the dedicated CNN architecture are key contributions.
• The findings support the application of this technology across various economic sectors requiring advanced object detection.