Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Visual Geo-Localization from images

Rania Saoud, Slimane Larabi

TL;DR

This work tackles GPS-denied geo-localization from images by integrating SIFT-based place recognition, VGG16-based road-junction classification, and graph-based map matching. It combines panoramic data generation, Detectron2 filtering, and a voting-based place anchoring scheme with a DL junction classifier trained via augmentation and SMOTE, all deployed in an offline mobile app. The approach yields high accuracy and robust performance across junction types, validated on a smartphone-collected dataset and supported by a GIS-driven map-matching framework. The results demonstrate practical viability for location inference in environments where traditional GPS signals are unavailable.

Abstract

This paper presents a visual geo-localization system capable of determining the geographic locations of places (buildings and road intersections) from images without relying on GPS data. Our approach integrates three primary methods: Scale-Invariant Feature Transform (SIFT) for place recognition, traditional image processing for identifying road junction types, and deep learning using the VGG16 model for classifying road junctions. The most effective techniques have been integrated into an offline mobile application, enhancing accessibility for users requiring reliable location information in GPS-denied environments.

Visual Geo-Localization from images

TL;DR

This work tackles GPS-denied geo-localization from images by integrating SIFT-based place recognition, VGG16-based road-junction classification, and graph-based map matching. It combines panoramic data generation, Detectron2 filtering, and a voting-based place anchoring scheme with a DL junction classifier trained via augmentation and SMOTE, all deployed in an offline mobile app. The approach yields high accuracy and robust performance across junction types, validated on a smartphone-collected dataset and supported by a GIS-driven map-matching framework. The results demonstrate practical viability for location inference in environments where traditional GPS signals are unavailable.

Abstract

This paper presents a visual geo-localization system capable of determining the geographic locations of places (buildings and road intersections) from images without relying on GPS data. Our approach integrates three primary methods: Scale-Invariant Feature Transform (SIFT) for place recognition, traditional image processing for identifying road junction types, and deep learning using the VGG16 model for classifying road junctions. The most effective techniques have been integrated into an offline mobile application, enhancing accessibility for users requiring reliable location information in GPS-denied environments.
Paper Structure (13 sections, 8 figures, 2 tables, 1 algorithm)

This paper contains 13 sections, 8 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed System
  4. Place Recognition SIFT descriptor
  5. Road junction Identification using deep learning
  6. Integration and Map Matching
  7. Experimental results
  8. Dataset Preparation and Annotation
  9. Data Splitting and Balancing
  10. Data Augmentation
  11. Training and Fine-Tuning of the VGG16 Model
  12. Results
  13. Conclusion

Figures (8)

  • Figure 1: Panoramic dataset taken in our university
  • Figure 2: Example of the selected images for place matching using detectron2 classification
  • Figure 3: Visual representation of the most similar place
  • Figure 4: Junction types: X-junctions (red), Y-junctions (yellow), T-junctions (Blue) and crossroads
  • Figure 5: The images represent the road intersections encountered along the route. These include various junctions such as T-junctions, Y-junctions, and crossroads. The images provide a visual representation of the real-world path.The path computed by our algorithm is indicated by purple dots which correspond to the real path images.
  • ...and 3 more figures