Intersection Scene Graphs for Traffic Signal-Lane Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for determining traffic signal and lane associations at signalized intersections are resource-intensive and costly to scale, relying on high-definition maps that require manual annotation and computationally intensive processing.
Innovation Solution
An end-to-end, learning-based system that uses machine learning models to detect keypoints and linkages in images of signalized intersections, generating a scene graph to determine traffic signal states and lane associations, allowing for scalable and efficient navigation of autonomous vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-definition maps with manual annotation are used to determine traffic signal and lane associations, then measurement precision is improved, but device complexity and loss of time increase
Solution Approach 1:
The patent replaces manual annotation processes with automated machine learning models that process images to detect traffic signals, lanes, and their associations. The system uses computer vision algorithms to automatically identify and associate traffic signals with corresponding lanes, eliminating the need for manual HD map creation and reducing system complexity while maintaining high precision.
Solution Approach 2:
The system creates simplified representations (scene graphs) that copy only the essential elements and relationships needed for navigation - traffic signals, lanes, and their associations - rather than using complete manually-annotated HD maps. This reduces data complexity while preserving the critical information needed for accurate traffic signal and lane association.
2Measurement precision
If high-definition maps with manual annotation are used to determine traffic signal and lane associations, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The machine learning models are pre-trained on large datasets of traffic scenes, enabling them to quickly and accurately detect traffic signals, lanes, and their associations without requiring time-consuming manual annotation during deployment. The system performs preliminary learning during the training phase, so that during actual operation, it can rapidly process new scenes and determine associations in real-time.
Solution Approach 2:
The system replaces the time-intensive manual annotation process with automated computer vision processing. The machine learning models automatically detect and associate traffic signals with lanes by processing images in real-time, eliminating the manual labor required to create and update HD maps, thereby significantly reducing the time loss associated with map creation and maintenance.
3Reliability
If existing systems are used to determine traffic signal and lane associations, then reliability is improved, but productivity decreases
Solution Approach 1:
The machine learning-based system provides a universal solution that can determine traffic signal and lane associations across diverse intersections and geographic locations without requiring location-specific manual annotation. The same trained models can be deployed universally to handle various traffic scenarios, improving both reliability through consistent performance and productivity through easy scaling to new locations.
Solution Approach 2:
The system uses learned patterns and relationships from training data to create simplified scene graphs that capture the essential associations between traffic signals and lanes. These copied representations can be rapidly generated for new intersections without manual intervention, enabling reliable and scalable deployment across multiple locations while maintaining association determination accuracy.
Data Source
AI summary
Systems, methods, computer-readable media, techniques, and methodologies are disclosed for performing end-to-end, learning-based keypoint detection and association. A scene graph of a signalized intersection is constructed from an input image of the intersection. The scene graph includes detected keypoints and linkages identified between the keypoints. The scene graph can be used along with a vehicle's localization information to identify which keypoint that represents a traffic signal is associated with the vehicle's current travel lane. An appropriate vehicle action may then be determined based on a transition state of the traffic signal keypoint and trajectory information for the vehicle. A control signal indicative of this vehicle action may then be output to cause an autonomous vehicle, for example, to implement the appropriate vehicle action.