AVPS Marker Recognition Using Partitioned Binarization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional marker detection methods for automated valet parking systems (AVPS) face challenges in accurately recognizing fiducial markers in diverse parking lot environments with uneven lighting conditions and distorted images, leading to poor localization performance.
Innovation Solution
A method and apparatus for recognizing AVPS markers using a multi-step image processing approach that includes normalizing the marker image, detecting a region of interest, generating multiple binarized images through partitioning and integration, and decoding the marker ID to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If general image processing methods with simple otsu threshold are used for binarization, then the processing is simple and fast, but the binarization performance deteriorates in environments with uneven lighting
Solution Approach 1:
The patent divides the image into multiple local regions and performs adaptive thresholding on each region separately. This segmentation approach allows the binarization to adapt to local lighting variations while maintaining overall processing efficiency, resolving the contradiction between simplicity and performance in uneven lighting conditions.
Solution Approach 2:
The patent applies different thresholding strategies to different regions of the image based on local characteristics. By making the binarization process locally adaptive rather than globally uniform, the system achieves high binarization performance across varying lighting conditions while keeping the overall approach computationally efficient.
2Reliability
If deep learning methods are used for marker detection, then occluded markers can be recognized well, but the marker is detected in the form of a box making it difficult to recognize in highly distorted images
Solution Approach 1:
The patent uses asymmetric shape context features that are sensitive to the geometric distortion of marker shapes. This asymmetric feature extraction approach enables accurate recognition of highly distorted markers by capturing their specific geometric transformations, while the occlusion handling comes from the robustness of the feature matching process.
Solution Approach 2:
The patent transforms the marker recognition problem by changing parameters from standard bounding box detection to shape context feature matching. This parameter transformation allows the system to handle both occluded markers and highly distorted markers effectively by focusing on geometric invariants rather than rigid bounding boxes.
3Productivity
If a single binarization method is used, then the processing is fast, but the accuracy of coded marker recognition is reduced in diverse environmental conditions
Solution Approach 1:
The patent implements a multi-stage binarization process that applies different binarization methods in sequence or in parallel, then integrates the results. This periodic application of different processing strategies maintains high accuracy across diverse conditions while managing computational load through structured processing stages.
Solution Approach 2:
The patent combines multiple binarization results through integration, merging the strengths of different binarization methods. This combination approach achieves high recognition accuracy in diverse environmental conditions while maintaining reasonable processing speed by leveraging the complementary nature of different binarization techniques.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The present disclosure provides a method for recognizing a coded marker including: obtaining a marker image including a coded marker from an input image and normalizing the marker image to obtain a normalized image detecting a region of interest in the normalized image; generating a first binarized image by performing binarization to classify each pixel of the normalized image into one of a foreground and a background; generating a second binarized image by dividing the normalized image into a plurality of partitions, performing binarization for pixel values for each partition to generate a binarized image for the each partition, and combining the binarized images for the each partition; generating an integrated binarized image according to a result of comparing pixel values at the same location in the first binarized image and the second binarized image; and obtaining identification data of the coded marker from the integrated binarized image.