Traffic Light Detection Auto-Labeling via V2I and Federated Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning approaches for traffic light detection rely heavily on large-scale annotated datasets, which are expensive to collect and curate, and require significant data transmission costs, making it challenging to train effective traffic light recognition models for autonomous vehicles.
Innovation Solution
The method employs vehicle-to-infrastructure (V2I) communications to aggregate traffic light signals at intersections, enabling auto-labeling of image training data and federated learning, where connected vehicles collaboratively train a shared traffic light prediction model onboard, reducing data transmission and preserving privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large-scale annotated datasets are used for traffic light detection, then detection performance is improved, but data transmission costs and storage requirements increase significantly
Solution Approach 1:
The system enables vehicles to autonomously generate and label training data through their own sensors and V2I communications. Each vehicle serves itself by capturing images, receiving traffic light state information from infrastructure, and automatically creating labeled training datasets without requiring centralized data collection and transmission
Solution Approach 2:
The training data generation process is segmented and distributed across multiple vehicles rather than centralized. Each vehicle independently generates its own labeled training data locally using its camera and V2I communications, dividing the overall data collection task into many small autonomous units that eliminate the need for large-scale data transmission
2Device complexity
If centralized data storage and processing are used, then model training is simplified, but privacy concerns and data transmission requirements increase
Solution Approach 1:
The centralized training system is segmented into distributed edge computing nodes (vehicles). Each vehicle performs local model training using its own generated training data, and only model parameters or gradients are shared via federated learning, keeping raw data private and localized while still achieving collaborative model improvement
Solution Approach 2:
Federated learning serves as an intermediary mechanism that enables collaborative model training without direct data sharing. The system uses communication protocols to exchange only model updates between vehicles and a central coordinator, acting as an intermediary that preserves data privacy while achieving the benefits of centralized training
3Measurement precision
If manual annotation of training data is performed, then data quality is improved, but time and cost requirements increase significantly
Solution Approach 1:
The system performs self-annotation by automatically generating training data with ground truth labels through V2I communications. Traffic light states are obtained directly from infrastructure controllers, providing accurate automated labels without human intervention, thus maintaining data quality while eliminating annotation time and cost
Solution Approach 2:
The manual mechanical process of human annotation is replaced with an automated electronic system. V2I communication protocols and sensor fusion algorithms automatically generate labeled training data by correlating camera images with traffic light state information from infrastructure, substituting human labor with automated computational processes
Data Source
AI summary
A method for traffic light auto-labeling includes aggregating vehicle-to-infrastructure (V2I) traffic light signals at an intersection to determine transition states of each driving lane at the intersection during operation of an ego vehicle. The method also includes automatically labeling image training data to form auto-labeled image training data for a traffic light recognition model within the ego vehicle according to the determined transition states of each driving lane at the intersection. The method further includes planning a trajectory of the ego vehicle to comply with a right-of-way according to the determined transition states of each driving lane at the intersection according to a trained traffic light detection model. A federated learning module may train the traffic light recognition model using the auto-labeled image training data during the operation of the ego vehicle.