Street-View Azimuth Estimation for Accurate Geo-Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the orientation and geo-location of street-view images are inaccurate, leading to misalignment in map creation and navigation, especially in areas lacking professional imagery, and crowdsourced images often lack orientation information.
Innovation Solution
A machine learning model using a weighted soft-margin triplet loss and absolute angle error loss functions is trained to estimate the azimuth rotation and geolocation of street-view images, employing a south-aligned orientation alignment coordinate system and fine-grained orientation estimation methods without additional learnable parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If cross-view geo-localization uses satellite imagery for estimating orientation and location, then the method can handle large-scale searches, but the accuracy of orientation estimation and geo-localization is insufficient
Solution Approach 1:
The patent changes the parameter space by introducing a normalized angle representation and using angle difference calculations instead of direct orientation angle estimation. This transforms the problem into estimating differences in orientation, which improves precision while maintaining large-scale search capability through the normalized coordinate system.
Solution Approach 2:
The patent replaces traditional mechanical image retrieval methods with a machine learning-based deep learning model that processes image features directly. This substitution enables more accurate orientation and location estimation by learning complex patterns from data rather than relying on simple image matching algorithms.
2Device complexity
If traditional orientation estimation methods are used, then the system remains simple, but the accuracy of geo-localization and orientation is insufficient for real-world applications
Solution Approach 1:
The patent segments the orientation estimation problem into multiple components: first estimating the coarse orientation using image retrieval, then refining it by calculating angle differences between corresponding features. This segmentation allows the system to maintain simplicity in the overall architecture while achieving high precision through the refined estimation step.
Solution Approach 2:
The patent introduces an intermediary representation using normalized angle coordinates and angle differences as a mediator between the image retrieval process and the final orientation estimation. This intermediary layer enables precise geo-localization while keeping the overall system architecture manageable and simple.
Data Source
AI summary
A computer assisted method comprising: storing a training dataset including a plurality of geotagged candidate images and a plurality of query images, each query image having at least one corresponding candidate image having the same geolocation; applying a quasi-random or random azimuth rotation to each of the plurality of query images, and storing the azimuth rotation for each of the plurality of rotated query images; training a machine learning model, including: extracting features from the plurality of rotated query images; estimating the azimuth rotation of the rotated query image based on an inference of the extracted features of the rotated query image and extracted features from the candidate images, and using an objective function including a first loss function based on a weighted soft-margin triplet loss, and a second loss function based on an absolute angle error between the stored azimuth rotation and the estimated azimuth rotation for the stored dataset.