AVPS Marker Detection Using Multi-Head Vision Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting AVPS markers, such as ArUco markers, face challenges with small marker detection, incorrect detection, and low recognition rates, especially in highly distorted images, which affect the accuracy of vehicle localization in automated valet parking systems.
Innovation Solution
A method and apparatus using a deep learning network trained through multi-task learning to simultaneously detect the bounding box and corner points of coded markers, involving image preprocessing, extraction of feature maps, and modified non-maximum suppression to enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If general image processing techniques or deep learning are used to detect ArUco markers, then the detection process can be implemented, but small markers cannot be detected, markers may be incorrectly detected, or marker recognition rate is very low for highly distorted images
Solution Approach 1:
The detection task is segmented into multiple sub-tasks handled by specialized heads: center point detection, bounding box regression, and corner point detection. Each head processes specific features independently, allowing optimized detection for different marker characteristics including small and distorted markers.
Solution Approach 2:
Different parts of the network are designed with specialized characteristics: the backbone network extracts general features, while each head network is optimized for its specific detection task. The center point head uses one type of feature extraction optimized for marker centers, while corner point heads use another optimized for corner detection, allowing each component to have optimal local quality for its function.
2Extent of automation
If vision-based localization using high definition map and cameras is adopted, then vehicle localization can be achieved, but the accuracy of coded marker recognition directly affects localization performance
Solution Approach 1:
The patent replaces traditional mechanical/algorithmic marker detection with a deep learning-based neural network system. The network automatically learns optimal feature representations and detection patterns from training data, substituting manual algorithm design with data-driven learning to achieve more accurate and robust marker recognition.
Solution Approach 2:
The network dynamically adjusts detection parameters through learned transformations. The offset head network learns to correct position errors by adjusting coordinate parameters, and the corner point head learns geometric relationships through parameter transformation, allowing the system to adapt to various marker conditions and maintain high localization accuracy.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An apparatus and a method for detecting an AVPS marker. In one example, a method of detecting a coded marker comprises receiving an image and preprocessing the image, extracting a feature map of the image through a backbone network, extracting a center point heatmap for a coded marker, a first feature map for a width and height of a bounding box, a second feature map for offsets for adjusting the center point, and a third feature map for corner points of the coded marker from the feature map using a plurality of head networks, generating candidate detection information including the bounding box and the corner points on the basis of the center point heatmap, the first feature map, the second feature map, and the third feature map, and outputting final detection information by performing modified non-maximum suppression on the candidate detection information.