Visual Docking Guidance for New Aircraft Type Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing aircraft docking systems rely heavily on human intervention and skilled ground marshallers, are time-consuming, and lack sufficient data sets for accurate guidance across various aircraft models, especially in adverse weather conditions, leading to inefficiencies and safety risks.
Innovation Solution
A learning-based guidance system that utilizes neural networks to automatically detect and classify aircraft types in real-time, generating synthetic docking guidance using low-cost image sensors and 3D modeling, capable of handling new aircraft models through synthetic video generation and semi-supervised learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Visual Docking Guidance System (VDGS) with sensors and dashboard is deployed, then aircraft docking precision and safety are improved, but system complexity and cost increase
Solution Approach 1:
The patent uses synthetic video generation to create virtual representations of aircraft docking scenarios. Instead of requiring extensive real-world video data for every aircraft type, the system generates synthetic copies of docking videos through 3D modeling and rendering, which are then used to train the deep learning model. This approach maintains high docking precision while reducing the complexity of data collection and system deployment.
Solution Approach 2:
The system performs preliminary actions by pre-generating synthetic training data for various aircraft types and conditions before actual docking operations. The deep learning model is pre-trained on these synthetic videos, enabling it to accurately detect and guide new aircraft types without requiring extensive real-world training data collection during deployment.
2Productivity
If ground marshallers are used to guide aircraft, then system simplicity is maintained, but time consumption and human error increase
Solution Approach 1:
The system enables self-service by implementing automated aircraft type detection and docking guidance through deep learning models. The system independently identifies aircraft types from video feeds, retrieves appropriate docking parameters, and provides guidance without requiring human marshallers, thereby eliminating time losses associated with manual intervention and reducing human error.
Solution Approach 2:
The patent replaces the mechanical system of ground marshallers with an automated electronic system based on deep learning and video analysis. The system uses computer vision to detect aircraft, processes video data through neural networks, and provides automated guidance displays, substituting human physical guidance with an electronic automation system that operates continuously without fatigue or error.
3Measurement precision
If sufficient real aircraft video data is collected for training, then detection accuracy improves, but data collection time and cost increase significantly
Solution Approach 1:
The patent employs synthetic video generation to create training data copies. Instead of collecting extensive real-world videos of every possible aircraft type, the system generates synthetic representations using 3D models and rendering engines. These synthetic videos capture various aircraft types, lighting conditions, and weather scenarios, providing sufficient training data for high detection accuracy without the time-consuming process of real-world data collection.
Solution Approach 2:
The system utilizes parameter changes by varying conditions in synthetic video generation, such as lighting, weather, camera angles, and aircraft positions. This allows the training model to learn robust feature recognition across diverse scenarios without requiring separate real-world video collections for each condition, maintaining high detection accuracy while minimizing data collection time.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Automated visual docking guidance in and near a bridge area is described herein. One method for aircraft detection, including capturing camera image data of a new aircraft; generating a segmented aircraft mask; segmenting the image data of the new aircraft into body part segmentation data; classifying the body part segmentation data into a plurality of classes; analyzing each class of body part segmentation data to predict an aircraft type for the new aircraft; determining the aircraft type of the new aircraft based on the prediction analysis; and synthetic video generation of new aircraft for generating aircraft specific docking guidance for the new aircraft based on the determined aircraft type.