Traffic Light Detection Using HD Map Pose-Guided Image Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for traffic light detection in autonomous vehicles require full image processing of the vehicle's front, which is computationally intensive and inefficient, especially in complex environments with multiple lanes and intersections.

Innovation Solution

A method and system that utilizes first and second pose information from a positioning module and high-definition map module to determine the position of a target traffic light in an image, allowing for focused image processing on a specific area centered on the traffic light, reducing computational complexity by eliminating the need for full image matching with HD MAP data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If full image processing is performed to detect traffic lights, then detection accuracy is maintained, but computational complexity increases significantly

Engineering Contradiction:
Improvetraffic light detection accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the image processing task into two stages: first, use pose information to identify and segment the region containing the target traffic light; second, perform detailed image processing only on this segmented region. This segmentation approach maintains detection accuracy while significantly reducing the computational load by avoiding full-image processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary action by using pose information from positioning modules and high-definition maps to pre-identify the location and coordinates of the target traffic light before image processing. This preliminary localization step guides subsequent image processing to focus only on the relevant region, reducing computational complexity while maintaining accuracy.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If full image processing is performed, then traffic light detection is reliable, but processing time increases

Engineering Contradiction:
Improvetraffic light detection reliabilityVSAvoidimage processing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

By segmenting the image processing task to focus only on the region containing the target traffic light (identified through pose information), the patent reduces processing time while maintaining detection reliability. The segmentation ensures that critical processing resources are concentrated on the relevant area only.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing complete and rigorous image processing only on the partial region that contains the target traffic light, rather than processing the entire image. This partial processing approach maintains reliability for the target object while significantly reducing overall processing time.

Inventive Principle:
Principle #16Partial or excessive action

3Measurement precision

If full image matching with HD MAP is performed, then positioning accuracy is improved, but computational load increases

Engineering Contradiction:
Improvepositioning accuracyVSAvoidcomputational load
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent uses pose information from positioning modules and high-definition maps to pre-determine the coordinates and location of the target traffic light before image processing. This preliminary action provides accurate positioning guidance without requiring computationally intensive full-image matching, thus maintaining positioning accuracy while reducing computational load.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent extracts and utilizes specific pose information (position and orientation data) from the positioning module and high-definition map to directly determine traffic light location. This extraction approach bypasses the need for full image-HD MAP matching, reducing computational load while preserving positioning accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4064220B1Method, system and device for detecting traffic light for vehicle
Publication Date: 2025.12.17 BAYERISCHE MOTOREN WERKE AG
  • EP4064220B1 patent drawingFigure 1~2
  • EP4064220B1 patent drawingFigure 3A~3B
  • EP4064220B1 patent drawingFigure 3C~3D

AI summary

The present disclosure relates to a method for detecting a traffic light for a vehicle, comprising: acquiring a first pose information for the vehicle from a positioning module, acquiring a second pose information for a target traffic light from a high-definition map module, and acquiring an image that presents an environment in front of the vehicle from a camera mounted on the vehicle; determining a position of the target traffic light in the image according to the first pose information and the second pose information; and performing image processing on an image area of the image that is located at the position so as to detect an indication of the target traffic light. The present disclosure further relates to a system and a device for detecting a traffic light for a vehicle.