A Sensor Fused Rear Cross Traffic Detection System Using Transfer Learning.

Area of Science:

• Automotive Engineering
• Computer Vision
• Sensor Fusion

Background:

• Advanced Driver Assistance Systems (ADAS) are crucial for improving road safety by mitigating collisions.
• Rear-end collisions are frequent, and a Rear Cross Traffic (RCT) detection system is vital for safety when reversing.
• Existing systems may have limitations in processing speed and accuracy for real-time obstacle detection.

Purpose of the Study:

• To propose a robust sensor-fused RCT detection system.
• To enhance the safety and efficiency of ADAS applications.
• To reduce processing time and improve object detection accuracy in reversing scenarios.

Main Methods:

• Developed a sensor fusion algorithm combining data from two radars and a wide-angle camera.
• Utilized a transferred Convolution Neural Network (CNN) model for object classification.
• Integrated sensor data to accurately identify target object locations and types.

Main Results:

• The proposed sensor-fused RCT system achieved 96.42% accuracy in object detection.
• Processing time was reduced by 15.34 times compared to a camera-only system.
• Demonstrated robust performance in identifying obstacles during reversing maneuvers.

Conclusions:

• The sensor-fused RCT detection system offers significant improvements in both accuracy and processing speed.
• This technology is vital for the practical deployment of advanced ADAS features.
• The system enhances driver safety by providing reliable rearward obstacle detection.