Explore visual geo-localization and location-aware image understanding in this 52-minute conference talk presented by Amir Zamir from the University of Central Florida. Dive into ground level imagery and geo-localization techniques using image matching. Examine various approaches to visual geo-localization, including geospatial pruning, quantitative localization results, and the use of multiple nearest neighbors. Learn about forming input graphs for Generalized Minimum Clique Problems and robustifying distance functions. Discover segment geo-localization methods, including MST-based trajectory reconstruction and triplet location estimation. Investigate visual business recognition techniques, multi-hypotheses character recognition, and the fusion of text processing with image matching. Gain insights into future directions for research in this field.
Visual Geo-Localization and Location-Aware Image Understanding
University of Central Florida via YouTube
Overview
Syllabus
Intro
Ground Level Imagery
Geo-localization Using Image Matching
Approaches to Visual Geo-localization
Dissertation
Outline
Chapter 3
Geospatial Pruning
Quantitative Localization Results
Ambiguity of local features
Using Multiple Nearest Neighbors
Forming Input Graph to GMCP
K-partite Complete Input Graph
Generalized Minimum Clique
Multiple group of matching images
Robustifying the distance function
Local Search
Geo-localization Results
Robustification Function
Chapter 5
Segment Geo-Localization
MST-based Trajectory Reconstruction
MST Based reconstruction results
Quantitative Results
Error in GPS tags
Triplet Location Estimation
Random Walks on GPS estimations
Experiments
Adaptive Damping Factor Evaluation
Chapter 7
Visual Business Recognition
Multi-Hypotheses Character Recognition
Fusing Text Processing and Image Matching
Summary
Directions for Future Work
Taught by
UCF CRCV
