CrosswalkNet: An Optimized Deep Learning Framework for Pedestrian Crosswalk Detection in Aerial Images with High-Performance Computing

2025-06-09

AI Summary

This work addresses the challenge of accurately detecting and vectorizing arbitrarily oriented pedestrian crosswalks (zebra crossings) in high-resolution (15 cm) aerial imagery. We propose CrosswalkNet, an end-to-end deep learning framework built upon an enhanced YOLOv8 architecture. Our method introduces oriented bounding box (OBB) detection to explicitly model crosswalk orientation, integrates Convolutional Block Attention Module (CBAM) for adaptive feature weighting, employs a dual-branch SPP-Fast module for multi-scale contextual aggregation, and adopts cosine annealing for robust optimization. The framework enables GPU-accelerated inference and outputs GIS-compatible polygon vectors (Shapefile format). Evaluated on the Massachusetts dataset, CrosswalkNet achieves 96.5% precision and 93.3% recall. Notably, it demonstrates strong zero-shot cross-domain generalization—maintaining high performance when directly deployed on unseen datasets from New Hampshire, Virginia, and Maine—enabling real-time traffic asset management and safety analytics.

Technology Category

• Computer Vision: Object Detection & Categorization
• Application Domains: Transportation
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

With the increasing availability of aerial and satellite imagery, deep learning presents significant potential for transportation asset management, safety analysis, and urban planning. This study introduces CrosswalkNet, a robust and efficient deep learning framework designed to detect various types of pedestrian crosswalks from 15-cm resolution aerial images. CrosswalkNet incorporates a novel detection approach that improves upon traditional object detection strategies by utilizing oriented bounding boxes (OBB), enhancing detection precision by accurately capturing crosswalks regardless of their orientation. Several optimization techniques, including Convolutional Block Attention, a dual-branch Spatial Pyramid Pooling-Fast module, and cosine annealing, are implemented to maximize performance and efficiency. A comprehensive dataset comprising over 23,000 annotated crosswalk instances is utilized to train and validate the proposed framework. The best-performing model achieves an impressive precision of 96.5% and a recall of 93.3% on aerial imagery from Massachusetts, demonstrating its accuracy and effectiveness. CrosswalkNet has also been successfully applied to datasets from New Hampshire, Virginia, and Maine without transfer learning or fine-tuning, showcasing its robustness and strong generalization capability. Additionally, the crosswalk detection results, processed using High-Performance Computing (HPC) platforms and provided in polygon shapefile format, have been shown to accelerate data processing and detection, supporting real-time analysis for safety and mobility applications. This integration offers policymakers, transportation engineers, and urban planners an effective instrument to enhance pedestrian safety and improve urban mobility.

Problem

• Detects pedestrian crosswalks in aerial images using deep learning
• Improves detection precision with oriented bounding boxes (OBB)
• Enables real-time analysis for urban safety and mobility

Innovation

• Uses oriented bounding boxes for precise detection
• Implements Convolutional Block Attention for optimization
• Leverages High-Performance Computing for real-time analysis