Communication Network Architectures for Driver Assistance Systems.

Area of Science:

• Computer Vision
• Automotive Safety
• Wireless Communication

Background:

• Growing importance of Autonomous Driver Assistance Systems (ADAS) for road safety.
• Need for advanced systems to detect and warn drivers about pedestrians, especially at crossroads.

Purpose of the Study:

• To propose and analyze an ADAS for detecting pedestrians crossing the road.
• To evaluate different communication architectures for timely alert dissemination.
• To define vehicle braking curves based on system latency for collision avoidance.

Main Methods:

• Utilizing a Convolutional Neural Network (CNN)-based Pedestrian Detector (PD) with local camera imagery.
• Implementing and analyzing three distinct communication architectures for alert forwarding.
• Evaluating PD performance using accuracy, precision, and recall metrics.
• Assessing communication solutions based on alert message forwarding time.

Main Results:

• CNN-based PD accuracy ranges from 70% to 100%, dependent on video resolution.
• Analysis of communication architectures highlights varying alert message forwarding times.
• Overall system latency (PD processing + communication) is quantified.

Conclusions:

• The developed ADAS effectively detects pedestrians and disseminates alerts.
• Communication architecture choice significantly impacts system responsiveness.
• Latency data is crucial for defining safety-critical vehicle braking curves to prevent pedestrian collisions.