Traffic Light Signal Detection Using Shape-Matching Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The detection of traffic light signals in autonomous or assisted driving systems is challenging due to their small size and complex background, leading to difficulties in accurate identification.
Innovation Solution
A method utilizing a deep neural network that extracts multiple layers of feature maps, selects layers with different scales, and applies a convolution layer with a kernel matching the traffic light shape, combined with positional confidence filtering and fusion of features from the red and green channels, to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional detection methods are used for traffic lights, then the detection process is simple, but the detection accuracy is low due to small target size and complex background
Solution Approach 1:
The patent segments the detection process into multiple stages: extracting feature maps from multiple layers of the deep neural network, selecting specific layers with different scales, applying shape-matching convolution kernels, and performing positional confidence filtering. This segmentation allows each stage to focus on specific aspects of the detection problem, improving overall accuracy while managing complexity systematically.
Solution Approach 2:
The patent introduces multi-scale feature maps from different layers of the deep neural network, adding a scale dimension to the detection process. By selecting feature maps with different scales and applying them to detect traffic lights of varying sizes, the system overcomes the limitation of single-scale detection and improves accuracy for small targets against complex backgrounds.
2Measurement precision
If multi-scale feature maps from multiple layers are used, then detection accuracy improves, but computational complexity increases
Solution Approach 1:
Instead of using all available feature maps from all layers of the deep neural network, the patent selectively uses feature maps from specific layers (e.g., seventh, tenth, and thirteenth layers of VGG network) that provide the most useful multi-scale information. This partial action approach achieves good detection accuracy while reducing unnecessary computational overhead from less relevant feature maps.
3Measurement precision
If shape-matching convolution kernels are applied, then detection precision improves, but processing time increases
Solution Approach 1:
The patent performs preliminary actions by first extracting and selecting multi-scale feature maps before applying the shape-matching convolution kernels. This preparation allows the convolution operation to focus directly on matching traffic light shapes in the pre-processed feature maps, improving detection precision while optimizing processing time by avoiding redundant computations.
4Reliability
If positional confidence filtering is applied, then false alarms are suppressed, but detection complexity increases
Solution Approach 1:
The patent implements feedback by using positional confidence filtering to evaluate and filter detection results. The system calculates positional confidence based on the detected traffic light position and image characteristics, then uses this feedback to suppress false alarms. This feedback mechanism improves reliability by eliminating incorrect detections while maintaining the overall detection process structure.
Data Source
AI summary
The present disclosure relates to a method, device, computer readable media, and electronic devices for identifying a traffic light signal from an image. The method for identifying a traffic light signal from an image includes extracting, based on a deep neural network, multiple layers of first feature maps corresponding to different layers of the deep neural network from the image. The method includes selecting at least two layers of the first feature maps having different scales from the multiple layers of the first feature maps. The method includes inputting the at least two layers of the first feature maps to a convolution layer having a convolution kernel matching a shape of a traffic light to obtain a second feature map. The method includes obtaining a detection result of the traffic light signal based on the second feature map.