Traffic Light Detection Using HD Map Pose-Guided Image Segmentation
Find Innovative SolutionsGenerate 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
Engineering Contradiction Analysis
1Measurement precision
If full image processing is performed to detect traffic lights, then detection accuracy is maintained, but computational complexity increases significantly
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.
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.
2Reliability
If full image processing is performed, then traffic light detection is reliable, but processing time increases
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.
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.
3Measurement precision
If full image matching with HD MAP is performed, then positioning accuracy is improved, but computational load increases
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.
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.
Data Source
Figure 1~2
Figure 3A~3B
Figure 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.