Skip to the main content

Original scientific paper

https://doi.org/10.17559/TV-20240720001866

DST-Based 3D Obstacle Detection and Classification for Autonomous Driving

Zhi Xiong Jin ; Geely University of China, 641423, Cheng Du China *
Ran Wen ; Geely University of China, 641423, Cheng Du China
Ji Lan Huang ; Geely University of China, 641423, Cheng Du China

* Corresponding author.


Full text: english pdf 1.058 Kb

page 1648-1659

downloads: 120

cite


Abstract

Accurate 3D obstacle detection and classification are crucial for autonomous driving systems. This study proposes a novel algorithm combining Dempster-Shafer Theory (DST) with deep learning techniques to address limitations in traditional computer vision methods for 3D scenes. This method combines the extended-corrosion algorithm based on depth information with multi-feature vector classification using back propagation neural networks. Experimental results on real-world datasets demonstrated an average detection accuracy of 95.81% and an effective obstacle recognition accuracy of 93.26%. The algorithm's average processing time of 191.10 ms per frame met real-time requirements for autonomous driving applications. This approach offers improved accuracy and robustness in complex 3D environments, advancing the field of obstacle detection for intelligent transportation systems.

Keywords

autonomous driving technology; classification and identification; DST; 3D obstacles; target detection

Hrčak ID:

335050

URI

https://hrcak.srce.hr/335050

Publication date:

30.8.2025.

Visits: 257 *