Crowdsourced Map Data Cleanup via Expectation-Maximization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current crowdsourcing platforms are inadequate for accurately locating targets of interest in images, as they either rely on ambiguous user inputs or fail to provide precise location data, leading to missed targets and false positives in applications like search and rescue or refugee assessment.
Innovation Solution
A platform that utilizes a crowdsourced search and locate server to process tagging data from multiple users, computing agreement and disagreement values, and performing an expectation-maximization analysis to estimate the precise locations of targets, thereby improving the accuracy of target identification and location.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If crowdsourced tagging data from multiple users is collected, then quantity of target location data increases, but measurement precision of target location decreases due to user ambiguity and errors
Solution Approach 1:
The system implements feedback loops where initial target location estimates are generated from crowd tagging data, then used to guide additional crowd members to specifically tag nearby targets. The system continuously refines location estimates by incorporating new tagging data and feedback from previous iterations, improving precision while utilizing the quantity of available data.
Solution Approach 2:
The system performs preliminary processing of crowd tagging data to generate initial target location estimates before using these estimates to guide subsequent data collection. This preliminary action allows the system to establish a baseline location that can then be refined through targeted additional tagging efforts.
2Productivity
If automated image analysis is used, then productivity of target identification increases, but reliability of target detection decreases due to lack of specialized training
Solution Approach 1:
The system merges automated image analysis capabilities with crowdsourced human tagging. Automated methods provide rapid initial processing and guide the distribution of crowd tasks, while human taggers provide reliable detection for complex or ambiguous targets. The combination leverages both speed and accuracy.
Solution Approach 2:
The system uses automated analysis results as an intermediary to guide and coordinate crowd tagging efforts. The automated system identifies regions of interest and suggested target locations, which then serve as intermediaries to focus human attention on specific areas needing verification or additional tagging.
3Ease of operation
If crowd members are randomly assigned image chips, then ease of operation increases, but measurement precision of target location decreases due to lack of focused verification
Solution Approach 1:
The system dynamically adjusts the task assignment strategy based on the current state of target location estimation. As precision requirements increase or uncertainty in certain regions is detected, the system transitions from random assignment to targeted assignment of image chips to crowd members near estimated target locations for verification.
4Measurement precision
If expectation-maximization analysis is performed iteratively, then measurement precision of target location improves, but loss of time increases due to multiple iterations
Solution Approach 1:
The system performs a limited number of expectation-maximization iterations rather than exhaustive iterations. This partial action provides sufficient precision improvement for most applications without incurring the full time cost of multiple iterations, balancing accuracy requirements with computational efficiency.
Data Source
AI summary
A system for large-scale crowd sourcing of map data cleanup and correction, comprising an application server that generates image data, sends image data to a user device, receives tagging data provided by the device user, and provides tags to a crowdsourced search and locate server based on tagging data from a user device, a crowdsourced search and locate server that receives tags from an application server, computes agreement and disagreement values and performs expectation-maximization analysis, and a map data server that stores and provides map data, and a method for estimating location and quality of a set of geolocation data.