Spatio-temporal Track Density Shaping for AI Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Collecting aircraft track trajectory data is a time-consuming process, and existing systems struggle to generate datasets with user-specified spatio-temporal density profiles, limiting the availability of realistic training data for air traffic control and AI training.
Innovation Solution
The system generates modified track datasets by optimizing initial synthetic or real datasets using mathematical optimization techniques, shifting and reshaping the data in time and space to meet desired density profiles, and employing a spatio-temporal coverage metric to control density.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If track data is collected through traditional methods, then data authenticity is maintained, but data collection time and resource consumption increase significantly
Solution Approach 1:
The patent creates synthetic copies of real track data through mathematical optimization and spatio-temporal transformations. Instead of collecting additional real data, the system generates realistic synthetic trajectories by transforming existing real data, maintaining authenticity characteristics while eliminating time-consuming collection processes.
Solution Approach 2:
The system pre-processes real track data to create a foundation dataset, then uses this preprocessed data to generate synthetic trajectories on demand. This preliminary action of preparing base data enables rapid generation of additional training data without repeated collection efforts.
2Quantity of substance
If more track data is collected to achieve desired density profiles, then data volume increases, but collection time and computational resources increase
Solution Approach 1:
The patent transforms existing data by changing spatio-temporal parameters through optimization. By adjusting parameters like track density profiles, time shifts, and spatial transformations, the system generates varied synthetic data from a fixed base dataset, increasing effective data volume without proportional increases in collection resources.
Solution Approach 2:
The synthetic data generation system serves multiple functions: it can create data with different density profiles, simulate various traffic conditions, and generate trajectories for different time periods all from the same base dataset, replacing multiple specialized data collection efforts with a single versatile system.
3Reliability
If real track data is used directly, then data realism is maintained, but spatio-temporal density control and customization are limited
Solution Approach 1:
The patent applies different transformations to different portions of the data to achieve desired local density characteristics. By selectively transforming specific spatio-temporal regions while preserving others, the system maintains realism in certain areas while achieving customized density control in targeted zones.
Solution Approach 2:
The system dynamically adjusts synthetic trajectory parameters based on desired density profiles. Rather than using fixed transformations, the optimization process adapts transformation parameters in real-time to match target density requirements, enabling flexible customization while preserving realistic motion patterns.
Data Source
AI summary
A system includes a first dataset of vehicle trajectories including individual vehicle tracks forming a first track density, a desired vehicle track density profile, a spatio-temporal coverage metric, and a track model including Heaviside functions encoding track time origins and durations for the individual vehicle tracks and including locations for the plurality of individual vehicle tracks. Approximations of the Heaviside functions are minimized as a function of the spatio-temporal coverage metric. The first dataset of vehicle trajectories is manipulated to satisfy the desired vehicle track density profile through minimizing an approximation of the Heaviside functions. Each individual vehicle track in the first dataset of vehicle trajectories is shifted as a result of the optimization, thereby generating a second dataset of vehicle trajectories having a second track density.