Scholarlink

We introduce the Intelligent Navigation System (INS), which addresses urban traffic congestion through computer vision and intelligent pathfinding. INS uses an adapted SAFECount model for real-time multi-type vehicle density estimation from public camera feeds, integrated with the A* algorithm for traffic-aware route optimization. Unlike conventional methods, INS estimates vehicle counts across categories (motorcycles, cars, buses, others) simultaneously in a single pass, using class-specific exemplars. We enhance model robustness through CutMix-inspired data augmentation and architectural optimizations. INS integrates these components into a web interface for route planning and monitoring, supported by a Django backend, a PostgreSQL database, and a RESTful API. Built with open-source technologies and public data, INS achieves reliable performance with inference times under 15 s across multiple camera feeds on consumer hardware.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Real-Time Traffic Density Estimation and Intelligent Navigation Using Few-Shot Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Quoc-Dat Pham-Le,
  • Bao-Nhi Duong-Ngoc,
  • Thanh-An Nguyen

摘要

We introduce the Intelligent Navigation System (INS), which addresses urban traffic congestion through computer vision and intelligent pathfinding. INS uses an adapted SAFECount model for real-time multi-type vehicle density estimation from public camera feeds, integrated with the A* algorithm for traffic-aware route optimization. Unlike conventional methods, INS estimates vehicle counts across categories (motorcycles, cars, buses, others) simultaneously in a single pass, using class-specific exemplars. We enhance model robustness through CutMix-inspired data augmentation and architectural optimizations. INS integrates these components into a web interface for route planning and monitoring, supported by a Django backend, a PostgreSQL database, and a RESTful API. Built with open-source technologies and public data, INS achieves reliable performance with inference times under 15 s across multiple camera feeds on consumer hardware.