Traffic Light Perception Using Virtual Lights Without HD Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current traffic light perception methods for autonomous driving rely on high-definition maps, which are prone to inconsistencies, leading to inaccurate traffic light status detection and increased vehicle safety risks.
Innovation Solution
A method that senses and recognizes the current image frame of a traffic light to determine the lamp head shape and color, forming a virtual light based on the traffic indication direction and color, allowing for accurate traffic light status detection without relying on pre-labeled high-definition maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-definition maps with pre-labeled traffic light information are used, then traffic light perception can be performed in covered regions, but the information becomes inconsistent with actual traffic lights when maps are not updated, leading to inaccurate detection
Solution Approach 1:
The patent creates a virtual light that copies the essential characteristics (position, color, shape) of the actual traffic light through perception model processing. This virtual representation serves as a dynamic copy that automatically updates with real-time traffic light status, eliminating the inconsistency problem of static map labels while maintaining accurate detection capability.
Solution Approach 2:
The patent transitions from static pre-labeled map information to dynamic real-time perception by continuously updating the virtual light based on current image frame analysis. The virtual light's position, color, and shape parameters are dynamically adjusted to match the actual traffic light state, ensuring both accuracy and consistency.
2Adaptability or versatility
If high-definition maps with pre-labeled traffic lights are used, then vehicle control can be performed based on labeled information, but the method cannot be performed in regions not covered by high-definition maps
Solution Approach 1:
The patent develops a universal perception method that works across all geographic regions regardless of high-definition map coverage. The perception model processes image frames to generate virtual lights for any traffic light in the camera's field of view, making the system adaptable to any location while maintaining detection accuracy through consistent image-based recognition.
Solution Approach 2:
Instead of relying on pre-labeled map data that requires geographic coverage, the patent creates virtual lights by copying visual information directly from image frames. This approach enables operation in any region where the camera can capture traffic lights, expanding geographic coverage while preserving detection accuracy through direct visual perception.
3Measurement precision
If 2D results from perception models are associated with projected traffic lights from high-definition maps, then traffic light color can be determined, but the association process becomes complex and error-prone when map information is inconsistent
Solution Approach 1:
The patent extracts traffic light information (position, color, shape) directly from the perception model's 2D detection results without requiring association with high-definition map projections. By taking out only the essential parameters needed for virtual light creation and eliminating the map association step, the system reduces complexity while maintaining color determination accuracy.
Solution Approach 2:
Instead of projecting map traffic lights onto the 2D image and matching with detection results, the patent inverts the process by directly using the 2D detection results to create the virtual light. This reversal eliminates the complex association and projection steps, simplifying the process while preserving accurate color determination from the perception model.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The disclosure relates to the field of autonomous driving, and specifically provides a traffic light perception method, a vehicle control method, a device, a medium, and a vehicle, to solve the problem of accurately sensing a status of a traffic light. For this purpose, the method provided in the disclosure includes: sensing and recognizing a current image frame of the traffic light to obtain a lamp head shape and a color of a currently lit single light in the traffic light; determining a traffic indication direction of a virtual light corresponding to the traffic light based on the lamp head shape of the single light; and determining a color of the traffic indication direction based on the color of the single light, to form a virtual light. According to the above method, a status of a traffic light can be accurately obtained without relying on a high-definition map, which improves the safety and reliability of vehicle driving.