Traffic Light Recognition Models for Diverse Signal Types
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep-learning models face challenges in effectively training and recognizing diverse types of traffic lights, including vehicle, pedestrian, and bicycle traffic lights, for autonomous vehicles in urban environments.
Innovation Solution
A traffic light recognition model learning and inference system that constructs a data set during autonomous driving, trains a deep-learning model using a traffic light data set, and infers traffic light recognition, utilizing a camera for image data collection, processors for sampling and annotation, and a deep-learning model for location and status inference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single deep-learning model is used for training, then the model structure is simple, but it cannot effectively recognize diverse types of traffic lights (vehicle, pedestrian, bicycle traffic lights)
Solution Approach 1:
The patent divides the traffic light recognition task into multiple specialized deep-learning models, each trained to recognize specific types of traffic lights (vehicle traffic lights, pedestrian traffic lights, bicycle traffic lights). This segmentation allows each model to specialize in detecting particular traffic light types, improving overall recognition accuracy across diverse traffic light scenarios without requiring a single overly complex model to handle all variations simultaneously.
2Measurement precision
If multiple deep-learning models are used for different traffic light types, then recognition accuracy improves, but system complexity increases
Solution Approach 1:
The system employs multiple specialized deep-learning models that are independently trained on specific traffic light types. Each model achieves high precision for its designated traffic light category, and the models can be selectively deployed based on the autonomous vehicle's operational context, balancing accuracy requirements with system complexity management.
Solution Approach 2:
While using multiple specialized models, the system maintains a unified architecture and training framework that can be adapted to different traffic light types. The data collection platform and processing pipeline serve multiple functions by collecting and processing data for various traffic light types through a common infrastructure, reducing overall system complexity despite the use of multiple recognition models.
3Adaptability or versatility
If comprehensive traffic light data is collected during autonomous driving, then the data set is diverse and representative, but data processing time and computational resources increase
Solution Approach 1:
The system performs preliminary actions by collecting and storing diverse traffic light data during autonomous driving operations in real-time. The data collection platform continuously captures images and metadata (location, timestamp, traffic light type) as the vehicle operates, building a comprehensive dataset beforehand. This preliminary data accumulation allows for efficient offline training and processing, reducing the computational burden and time required during model training and deployment phases.
Data Source
AI summary
Disclosed is a system for executing a traffic light recognition model learning and inference method, includes a data collection platform including a camera for collecting image data, and a first processor that samples traffic light image data including a traffic light among the image data, generates annotation data based on the traffic light image data, and generates a traffic light data set using the traffic light image data and the annotation data, wherein the traffic light data set includes information on a location of the traffic light, a type of the traffic light, traffic light on/off, and a traffic signal direction of the traffic light.