Front Vehicle Detection Algorithm for Smart Car Based on Improved SSD Model.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Driving Systems

Background:

• Conventional vehicle detection methods struggle with environmental variations, impacting accuracy and real-time performance.
• Deep learning-based object detection, particularly the Single Shot Multibox Detector (SSD), is a mainstream approach for smart car perception.

Purpose of the Study:

• To propose an improved SSD-based algorithm for accurate and real-time front vehicle detection in smart cars.
• To address limitations of existing methods in diverse environmental conditions and adverse weather.

Main Methods:

• Enhancements to the basic SSD network structure.
• Implementation of weighted masks during network training.
• Optimization of the loss function for improved detection performance.

Main Results:

• Achieved a mean Average Precision (mAP) of 92.18% on the KITTI dataset.
• Realized an average processing time of 15 milliseconds per frame.
• Demonstrated superior accuracy and real-time capabilities compared to existing deep learning methods.

Conclusions:

• The proposed algorithm offers simultaneous accuracy and real-time performance for front vehicle detection.
• Exhibits excellent robustness and adaptability in complex traffic environments and adverse weather.
• Significantly contributes to the safe and efficient operation of smart cars, potentially reducing traffic accidents.