Bulb-wise Traffic Light State Estimation for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional traffic light perception models are limited to detecting traffic lights with simple vertical or horizontal layouts and fail to accurately perceive alternative layouts or bulb configurations found in different regions, making them ineffective for universal use across various traffic light scenarios.
Innovation Solution
The method involves generating images of traffic lights, detecting characteristics of individual bulbs, temporally filtering these characteristics to refine accuracy, and outputting bulb-specific states, which are then used to generate autonomous vehicle navigation plans, allowing for universal detection of traffic lights regardless of layout or configuration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional traffic light perception models detect traffic lights with simple vertical or horizontal layouts, then detection accuracy is maintained for standard configurations, but the system fails to accurately perceive alternative layouts or bulb configurations found in different regions
Solution Approach 1:
The traffic light detection system segments the traffic light into individual bulb components, where each bulb is detected and classified separately. This segmentation allows the system to handle diverse traffic light layouts by treating each bulb as an independent detection target, thereby improving adaptability to different regional configurations while maintaining detection precision through component-level analysis
Solution Approach 2:
The system employs a universal detection framework that can handle multiple traffic light layouts and bulb configurations through a single unified model. The bulb-wise detection approach serves multiple functions: detecting presence, determining state, and classifying position, making the system versatile across different regional traffic light designs without sacrificing accuracy
2Measurement precision
If the system detects characteristics of each individual bulb, then accuracy of traffic light state perception is improved, but computational complexity and processing time increase
Solution Approach 1:
By segmenting the traffic light detection task into individual bulb detections, the system simplifies the overall problem into manageable atomic units. Each bulb is detected independently using the same detection head, which reduces the complexity of handling diverse layouts while maintaining high precision through consistent component-level analysis
Solution Approach 2:
The system uses a unified detection head that is copied and applied to each bulb independently. This copying approach allows the same detection logic to be reused across multiple bulbs, reducing overall system complexity while achieving high accuracy through consistent application of the detection model to each component
3Measurement precision
If temporal filtering is applied to refine detected bulb characteristics, then detection accuracy is enhanced, but processing time and computational load increase
Solution Approach 1:
The system applies temporal filtering at periodic intervals rather than continuously, processing bulb characteristic data at discrete time steps. This periodic application of filtering reduces computational load and processing time while still maintaining detection accuracy by capturing essential temporal patterns in traffic light state transitions
Data Source
AI summary
Systems and methods of traffic light detection/perception are provided for detecting traffic lights states and/or transitions for different traffic light layouts. Basic traffic light information can be obtained regarding, e.g., the existence of a traffic light(s) at an intersection being approached by a vehicle, and which bulb(s) may apply to/control a particular lane of traffic at the intersection. Based on that information, bulb-specific or bulbwise detection can be performed. Upon determining the existence of a traffic light, specific bulbs within or making up the traffic light can be detected, and their various states can be predicted or estimated relative to a corresponding bulb group. The various states can then be simplified and output as a message(s) upon which autonomous control of a vehicle can be based.