Real-Time Road Intersection Detection in Sparse Point Cloud Based on Augmented Viewpoints Beam Model.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Geospatial Data Analysis

Background:

• Road intersections are critical navigation landmarks but are challenging to detect robustly and efficiently in large-scale, sparse point clouds.
• Existing detection methods often struggle with the complexity and scale of real-world road networks.

Purpose of the Study:

• To propose a novel, real-time algorithm for accurate road intersection detection and localization in large-scale, sparse point clouds.
• To improve the robustness and efficiency of intersection detection compared to traditional approaches.

Main Methods:

• Development of an augmented viewpoints beam model to perceive road bifurcation structures.
• Joint extraction of spatial features from point clouds across multiple viewpoints.
• Integration of self-assessment evaluation metrics for real-time optimization of the detection process.
• Collection and annotation of a new VLP-16 point cloud dataset (NCP-Intersection) specifically for road intersections.

Main Results:

• The proposed method achieves an average precision exceeding 90% for road intersection detection.
• The algorithm demonstrates an average processing time of approximately 88 ms/frame, enabling real-time performance.
• Quantitative and qualitative experiments show favorable performance compared to existing methods.

Conclusions:

• The developed algorithm offers a robust and efficient solution for real-time road intersection detection in challenging point cloud data.
• The novel augmented viewpoints beam model and self-assessment metrics contribute to improved accuracy and speed.
• The NCP-Intersection dataset provides a valuable resource for future research in this domain.