Narrow Object Detection Using Merged Line Segments for Precise Cutting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network-based object detection algorithms struggle to accurately detect long, narrow objects like straps on cardboard boxes due to high background-foreground pixel ratios, leading to unstable cut point proposals for cutting stations in Automated Unpacking Machines.
Innovation Solution
A method involving gradient-based line segment detection, merging based on angular and spatial proximity, and filtering to accurately identify narrow objects, combined with object detection using YOLOv5 and cutting region proposals tailored to cutting machine capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If neural network-based object detection algorithms are used to detect straps on cardboard boxes, then the detection can be performed automatically, but the detection precision deteriorates due to high background-foreground pixel ratios causing noisy pixel and image features
Solution Approach 1:
The detection process is segmented into multiple stages: gradient calculation, line segment detection, line segment merging, and filtering. This multi-stage segmentation allows each stage to focus on specific features, progressively refining the detection from raw image data to precise strap localization, thereby resolving the contradiction between automation and precision.
Solution Approach 2:
Gradient-based line segment detection serves as an intermediary mechanism between the raw image input and the final object detection output. This intermediary step transforms the image into gradient information and line segments, which are more suitable for precise strap detection, thus improving detection precision while maintaining automation.
2Ease of operation
If gradient-based line segment detectors are used to detect narrow objects, then the detection can be performed, but the detection accuracy deteriorates because single pixel lines are detected twice and line segments are broken into smaller segments
Solution Approach 1:
The line segment merging step combines detected line segments that are spatially and angularly close into unified segments. This merging process corrects the fragmentation issue where single lines are broken into multiple segments, thereby improving localization accuracy while preserving the automated detection capability.
Solution Approach 2:
The filtering step uses feedback from the merging results to refine the final detection output. By evaluating the merged line segments against expected strap characteristics, the system corrects detection errors and improves localization accuracy, maintaining the balance between detection capability and precision.
3Device complexity
If rough localization of straps is used in Automated Unpacking Machines, then the processing can be simplified, but the cutting precision deteriorates leading to unstable cut point proposals
Solution Approach 1:
The detection pipeline is segmented into gradient calculation, line segment detection, merging, and filtering stages. This segmentation allows the system to maintain relatively simple processing at each stage while achieving high overall precision, resolving the contradiction between processing complexity and cutting precision.
Solution Approach 2:
The gradient calculation and line segment detection are performed as preliminary actions before the final filtering and cut point proposal. These preliminary steps prepare precise strap localization data in advance, ensuring that the cutting station receives accurate information without requiring complex real-time processing, thus maintaining both simplicity and precision.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables precise and efficient detection of narrow objects like straps, improving automation in unpacking processes by reducing false positives and enhancing cutting precision in Automated Unpacking Machines.
Implementation Method 1
Providing at least one image, the image comprising a representation of the narrow object, the image being a result of an acquisition by at least one sensor
Implementation Method 2
Determining gradient based line segments in the provided image, based on changes in brightness or colour intensity along the line. This may be achieved through a calculation of an image gradient, which particularly measures a rate of change of the brightness or colour intensity in different directions
Data Source
Figure 1~2
Figure 3
AI summary
The invention relates to a method (100) for detecting a narrow object (1), comprising the following steps: - Providing (101) at least one image, the image comprising a representation of the narrow object (1), the image being a result of an acquisition by at least one sensor (2), - Determining (102) gradient based line segments in the provided image, - Merging (103) the gradient based line segments based on an angular and a spatial proximity, - Filtering (104) the merged line segments to detect the narrow object (1). Furthermore, the invention relates to a computer program, an apparatus, and a storage medium for this purpose.