Traffic Signaling State Detection With Recurrent Neural Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing autonomous vehicles face challenges in accurately and efficiently detecting the states of traffic signaling units due to the variety of configurations and flashing lights, leading to delayed and inaccurate decision-making.

Innovation Solution

Employing a recurrent neural network, such as a long short-term memory (LSTM) neural network, to process sequences of images of traffic signaling units, combining them with pre-defined models to account for temporal dependencies and improve state detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If heuristically coded models are used to interpret traffic light signaling states, then the system can operate with simpler programming, but the response time is delayed due to waiting for extended time intervals to confirm light states

Engineering Contradiction:
Improveprogramming simplicityVSAvoidresponse time
Core Design Contradiction:
Ease of operationVSLoss of time

Solution Approach 1:

The system pre-processes and stores historical lighting state data and contextual information from other lighting elements before they are needed for decision-making. This preliminary preparation allows the neural network to quickly determine traffic light states without waiting for extended observation intervals, thus improving response time while maintaining operational simplicity

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system transitions from traditional time-based sequential analysis to a multi-dimensional approach by incorporating spatial context from multiple lighting elements and historical temporal patterns simultaneously. This dimensional expansion allows parallel processing of multiple data sources, reducing the time needed to confirm light states while keeping the system easy to operate

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Reliability

If heuristically coded models wait for extended time intervals to confirm light states, then false positives are reduced, but the detection accuracy is compromised by delayed responses to actual state changes

Engineering Contradiction:
Improvefalse positive reductionVSAvoidstate detection accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The system implements feedback mechanisms where the neural network continuously receives and processes historical lighting state data and contextual information from multiple lighting elements. This feedback loop allows the system to maintain high detection accuracy by constantly updating its understanding of traffic light states without requiring extended waiting intervals, while still filtering out false positives through pattern recognition

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

By pre-processing and storing historical data and contextual information beforehand, the system prepares the necessary evidence for accurate state detection in advance. This preliminary action enables the neural network to make accurate determinations quickly, reducing false positives without compromising detection precision or requiring delayed responses

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If the system incorporates context from other lighting elements and historical context, then detection accuracy improves, but the computational complexity increases

Engineering Contradiction:
Improvedetection accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The neural network acts as an intermediary that automatically processes and integrates complex contextual information from multiple lighting elements and historical data. This intermediary handles the computational complexity internally, providing accurate detection results without requiring the user or external system to manage the complexity directly, thus improving accuracy while managing computational burden

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The neural network performs self-service by automatically selecting and processing relevant contextual information and historical data without external intervention. This self-service capability allows the system to incorporate multiple data sources for improved accuracy while autonomously managing the computational complexity, reducing the burden on external systems

Inventive Principle:
Principle #25Self-service

4Speed

If traditional models process traffic light states without contextual information, then the processing is faster, but the response accuracy is reduced due to inability to distinguish flashing states

Engineering Contradiction:
Improveprocessing speedVSAvoidstate detection accuracy
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The system merges current lighting state observations with historical contextual information and data from other lighting elements into a unified neural network processing framework. This combination allows the system to maintain fast processing speeds through efficient neural network architecture while simultaneously improving accuracy by considering multiple contextual factors including flashing patterns that traditional models would miss

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4182839B1Detecting traffic signaling states with neural networks
Publication Date: 2025.09.03 WAYMO LLC
  • EP4182839B1 patent drawingFigure 1
  • EP4182839B1 patent drawingFigure 2A~2D
  • EP4182839B1 patent drawingFigure 3

AI summary

Machine-learning models are described detecting the signaling state of a traffic signaling unit. A system can obtain an image of the traffic signaling unit, and select a model of the traffic signaling unit that identifies a position of each traffic lighting element on the unit. First and second neural network inputs are processed with a neural network to generate an estimated signaling state of the traffic signaling unit. The first neural network input can represent the image of the traffic signaling unit, and the second neural network input can represent the model of the traffic signaling unit. Using the estimated signaling state of the traffic signaling unit, the system can inform a driving decision of a vehicle.