Dual-Camera Traffic Light Recognition Across Varying Distances
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle systems using a single camera for traffic light recognition face challenges with angle of field and effective distance, leading to inefficient detection of traffic light states, as the field of vision is either too narrow to capture distant lights or too wide to detect closer ones effectively.
Innovation Solution
The method employs a combination of a long-focus camera and a short-focus camera, along with a positioning sensor and high-precision map server, to obtain and recognize traffic light states by adaptively switching between the two cameras based on the vehicle's position and direction, ensuring accurate detection within a preset distance range.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a single camera with larger angle of field is used, then the field of vision is expanded to capture distant traffic lights, but the image size of traffic lights becomes smaller and detection difficulty increases
Solution Approach 1:
The patent divides the single camera system into two separate camera systems: a long-focus camera for capturing distant traffic lights and a short-focus camera for capturing close traffic lights. This segmentation allows each camera to be optimized for its specific detection range, resolving the contradiction between field of vision and image detection difficulty.
Solution Approach 2:
The patent introduces the dimension of focal length differentiation, creating a multi-dimensional camera system where cameras with different focal lengths (long-focus and short-focus) work together. This dimensional change allows the system to capture traffic lights at various distances with appropriate image sizes, solving the detection difficulty problem.
2Measurement precision
If a single camera with longer focal distance is used, then the image size of traffic lights is enlarged for better detection, but the viewing angle becomes smaller and traffic lights deviating from the field of vision cannot be collected
Solution Approach 1:
The patent segments the detection function into two specialized cameras: long-focus camera for distant traffic lights with appropriate image size, and short-focus camera for close traffic lights with enlarged images. This segmentation allows each camera to maintain optimal measurement precision for its specific range while collectively providing comprehensive field of vision coverage.
Solution Approach 2:
The patent creates a universal camera system where multiple cameras with different focal lengths work together to perform the universal function of detecting traffic lights at any distance. The system can adaptively select or combine images from different cameras based on the distance to the traffic light, achieving both precision and versatility.
3Device complexity
If only a single camera is used for traffic light recognition, then the device complexity is reduced, but the recognition efficiency and accuracy for traffic lights at various distances deteriorates
Solution Approach 1:
The patent segments the traffic light recognition task into two parallel processing streams: one for long-focus camera images and another for short-focus camera images. The system can independently process images from either camera based on the detected distance, maintaining efficient recognition while managing device complexity through modular architecture.
Data Source
AI summary
The method comprises: obtaining a first image collected by a long-focus camera, a second image collected by a short-focus camera, and positioning information and travelling direction of the autonomous vehicle collected by a positioning sensor at a target moment; according to the positioning information and traveling direction of the autonomous vehicle at the target moment, obtaining, from a high-precision map server, location information of traffic lights within a range of preset distance threshold ahead in the traveling direction of the autonomous vehicle at the target moment; recognizing the state of traffic lights at the target moment, according to the location information of the traffic lights at the target moment, and one image of the first image and second image.