Neural Network Container Alignment via Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing container alignment methods, particularly in filling and labeling systems, often rely on geometric features and image recognition, but can be inefficient and imprecise due to low match quality between actual and target image data, especially when the geometric feature is not aligned correctly with the camera axis.
Innovation Solution
A neural network is trained to recognize the geometric container feature in various orientations and distinguish between the search instance and exclude instances, enabling faster and more accurate alignment by sorting out recordings that do not contain the search instance, and using perspective distortion calculations to determine rotation adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image recognition methods are used to align containers, then the system can process containers through multiple alignment stages, but the alignment process becomes slow and imprecise when geometric features are not aligned correctly with the camera axis
Solution Approach 1:
The neural network is trained in advance with numerous example images of the geometric container feature in various orientations and lighting conditions. This preliminary training enables the network to immediately recognize and correctly identify the search instance during actual alignment operations, eliminating the need for multiple slow alignment stages and significantly reducing alignment time while improving precision.
Solution Approach 2:
The patent replaces traditional algorithm-based image recognition and comparison methods with a neural network-based system. This substitution enables faster and more accurate identification of the geometric container feature, allowing the system to quickly determine the correct alignment angle without relying on multiple sequential alignment stages, thus reducing both time and improving precision.
2Adaptability or versatility
If multiple cameras are used to record images of the container from different angles, then the system can search for the geometric feature in various orientations, but incorrect matches are selected when the search instance is not visible in certain camera views
Solution Approach 1:
The control electronics evaluate the match quality between recorded actual images and target image data for each camera. When a camera's recorded image shows a low match quality indicating the search instance is not visible or incorrectly oriented, the system identifies this as an exclude instance and discards it. This feedback mechanism ensures that only high-quality matches from cameras with proper viewing angles are used for alignment, maintaining both adaptability and precision.
Solution Approach 2:
The patent applies different evaluation criteria to different camera recordings based on their individual match quality. Each camera's image is independently evaluated, and only those meeting a quality threshold are considered valid search instances. This local quality assessment ensures that the system adapts to the specific viewing conditions of each camera while maintaining overall identification accuracy.
Data Source
Figure 1
AI summary
The invention relates to a method for aligning a container (2) in relation to at least one geometric container feature in a set position, wherein an alignment of the container (2) is carried out by an analysis of actual image data in a control circuit (12) taken by means of an image recording device (8, 9, 10, 11). The actual image data is analysed in a region for evaluation by means of pattern recognition, which provided an actual position of the container as a result if the geometric container feature is recognised and displays if the feature has not been recognised. The invention further relates to a device for carrying out said method and a computer programme product to carry out all the steps of the method.