AI Training Data Generation for Aircraft Runway Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in generating sufficient and reliable training data for deep learning algorithms to recognize runways, especially in degraded visibility conditions, due to the variability of conditions and the high cost of collecting real sensor images.
Innovation Solution
A method and device for automatically generating labeled training data using a flight simulator and sensor simulator to produce simulated sensor data, which is then used to create a ground truth, enabling the training of deep learning algorithms for runway recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real sensor images are collected to train deep learning algorithms, then the training data reflects real-world conditions, but the cost and time required for data collection becomes prohibitively high
Solution Approach 1:
The patent uses flight simulators to generate synthetic sensor images that copy real-world flight conditions and runway environments. These simulated images serve as training data substitutes, eliminating the need for expensive and time-consuming real sensor data collection while maintaining ecological validity through realistic simulation parameters
Solution Approach 2:
The system performs preliminary generation of diverse training scenarios including various weather conditions, lighting situations, and runway configurations before actual training is needed. This pre-generation approach allows researchers to have ready-to-use training datasets without waiting for real flight data collection
2Reliability
If diverse weather conditions are simulated to improve algorithm robustness, then the detection capability in degraded visibility improves, but the complexity of the simulation system increases
Solution Approach 1:
The flight simulator is configured to generate multiple types of sensor data (visual, infrared, millimetric radar) and simulate various weather conditions (fog, rain, snow) using a single multi-functional platform. This universal approach allows diverse training scenarios to be created without requiring separate specialized systems for each condition
Solution Approach 2:
The system varies simulation parameters such as visibility levels, precipitation intensity, lighting conditions, and runway surface states to create diverse training scenarios. By systematically changing these parameters, the system generates robust training data without adding physical complexity to the simulation infrastructure
3Productivity
If synthetic training data is generated using flight simulators, then the quantity of training data increases significantly, but the ecological validity of the training data may be reduced
Solution Approach 1:
The system uses the flight simulator's own capabilities to generate both the training data and the ground truth labels automatically. The simulator self-configures realistic flight paths, sensor parameters, and environmental conditions without requiring manual intervention, ensuring ecological validity while maintaining high productivity
Solution Approach 2:
The system incorporates feedback mechanisms where simulated sensor data is compared against ground truth from the same simulation to verify accuracy. This self-validation process ensures that synthetic training data maintains ecological validity by confirming that simulated observations match expected physical realities
Data Source
AI summary
A computer-implemented method for generating labeled training data for an artificial intelligence machine is provided. The method comprises at least the steps of: defining parameters for a scenario of an aircraft approaching a runway; using the parameters of the scenario in a flight simulator to generate simulated flight data, the flight simulator being configured so as to simulate the aircraft in the approach phase and an associated autopilot; using the simulated flight data in a sensor simulator to generate simulated sensor data, the sensor simulator being configured so as to simulate a forward-facing sensor on board an aircraft and able to provide sensor data representative of information of interest of a runway, the simulated sensor data that are generated being representative of information of interest of the runway; and using the simulated flight data and the simulated sensor data to generate a ground truth, the ground truth associated with the simulated sensor data forming a pair of simulated labeled training data.