Supervised Address-to-Building Matching for Accurate Map Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for matching building addresses with building outlines on maps suffer from low precision and coverage due to reliance on address geocodes, GPS inaccuracies, and unlabeled buildings, leading to increased delivery difficulties and delivery errors.
Innovation Solution
A supervised machine learning model that utilizes address and building data, including road segments and past delivery scans, to automatically assign positional orders to buildings and determine candidate buildings for accurate matching, using features like KDE and pairwise ranking to improve labeling precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional geocoding methods are used to match addresses with building outlines, then the process is simple and fast, but the matching precision is low and many buildings remain unlabeled
Solution Approach 1:
The patent introduces an intermediary supervised machine learning model that acts as a mediator between address geocodes and building outlines. This model processes multiple features including delivery scan points, building outlines, and address information to produce accurate matching results, thereby resolving the contradiction between simple geocoding and precise matching.
Solution Approach 2:
The patent replaces traditional mechanical/geometric matching methods with a supervised machine learning system. Instead of relying solely on coordinate-based proximity matching, the system uses trained models that process spatial relationships, building features, and delivery patterns to determine matches, significantly improving precision.
2Measurement precision
If GPS-based delivery scan points are used to identify building locations, then the method is easy to implement, but GPS inaccuracies cause geocodes to be located between neighboring buildings
Solution Approach 1:
The patent segments the delivery identification process into multiple components: collecting delivery scan points, generating building outlines from these points, and then using a machine learning model to match addresses with buildings. This segmentation allows each component to be optimized independently, improving overall accuracy while maintaining implementation feasibility.
Solution Approach 2:
The patent performs preliminary actions by collecting multiple delivery scan points over time and generating building outlines before the actual address matching process. This preliminary preparation creates more accurate spatial representations of buildings, which then serve as better inputs for the matching algorithm, reducing GPS accuracy issues.
3Quantity of substance
If heuristic methods are used to label buildings, then the process is straightforward, but the coverage of labeled buildings remains insufficient
Solution Approach 1:
The patent incorporates feedback mechanisms where the supervised machine learning model is trained on labeled data and then used to identify additional buildings. The system continuously improves by using accurate matches to refine the model, which then identifies more buildings with high confidence, creating a positive feedback loop that increases coverage while maintaining precision.
Solution Approach 2:
The patent changes key parameters of the labeling system by transitioning from simple heuristic rules to a supervised machine learning approach with multiple input features. This parameter change enables the system to process more complex patterns and relationships, significantly increasing the number of buildings that can be accurately labeled.
Data Source
AI summary
Systems and methods are provided supervised machine learning for matching addresses and buildings. Particularly, a machine learning model may be provided that is trained to identify a building outline within a map that is most likely associated with a given delivery address. A candidate set of buildings that may potentially be associated with the delivery address are determined and provided to the machine learning model. The machine learning model ranks the buildings included in the candidate set of buildings in terms of likelihood of being associated with the address. Based on this ranking, a building outline representing a building that is determined to be the most likely building associated with the address may be updated to include the address label on the map.