Object Configuration Recognition Using Orientation-Normalized Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for distinguishing different configuration states of an object, such as human gestures, are challenging due to limited computing power and the difficulty in applying basic feature descriptors under changing conditions.
Innovation Solution
A method involving image preprocessing, orientation determination, region-of-interest extraction, and histogram-based feature extraction in a normalized reference frame, which requires minimal computing power and is robust to object size and orientation changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple separate machine learning models are trained for each configuration state, then classification accuracy for each state is improved, but device complexity and computational resources increase
Solution Approach 1:
The patent combines multiple configuration-specific machine learning models into a single unified model. The model receives an image and a configuration state identifier as inputs, and outputs the defect classification. This merging approach maintains the ability to accurately classify defects across different configuration states while reducing the overall system complexity compared to having separate models for each state.
2Measurement precision
If configuration-specific training data is collected and trained separately, then training data accuracy is improved, but time consumption and processing speed decrease
Solution Approach 1:
The patent performs preliminary organization of training data by configuration states during the data preparation phase. The unified model is trained on a comprehensive dataset that includes labels indicating the configuration state for each defect example. This preliminary organization allows the model to learn configuration-specific defect patterns during a single training process, eliminating the need for multiple separate training runs and significantly reducing total training time.
3Adaptability or versatility
If separate models are used for different configuration states, then adaptability to specific configurations is improved, but system resource consumption increases
Solution Approach 1:
The patent implements a dynamic approach where a single unified model adapts its processing based on the input configuration state identifier. The model maintains configuration-specific knowledge through its training on diverse data, but dynamically adjusts its predictions based on the indicated configuration state. This dynamic adaptation allows the system to handle different configurations efficiently using one model, reducing computational resource consumption compared to loading and executing multiple separate models.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for distinguishing different configuration states of an object based on an image representation of the object comprises: Acquiring original image data representing an original image comprising a picture of the object; Detecting the object within the original image; Determining an orientation of the detected object within the original image; Determining within the original image a region-of-interest, ROI representing the picture of the object at least in parts; Generating normalized image data representing a two-dimensional normalized image by transforming the ROI of the original image based on the determined orientation of the object to a reference frame in which the ROI is represented by the normalized image with a predefined standard orientation and scaling; Extracting one or more characteristic features of the object as represented by the normalized image data; and Classifying the extracted one or more characteristic features according to a related classification scheme to determine among a set of two or more possible configuration states of the object a specific configuration state that is associated with the result of the classification. The method may be implemented by means of a computer program and/or an apparatus being configured to perform the method.