Image Processing Apparatus for Action Categorization Using Contour Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques face difficulties in easily determining the movement or state of a subject in a captured image due to the complexity of applying shape models to images with various edge shapes, making it challenging to categorize actions accurately.
Innovation Solution
The method involves acquiring direction information from captured images using a storage unit that associates shapes with contours, and a processor determines movement based on the relationship between this information, reducing the need for complex algorithms and simplifying the determination process by using an outline Hidden Markov Model (HMM) and event determination tables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If shape models are applied to images with various edge shapes to detect action patterns, then action categorization can be performed, but the determination process becomes complex and time-consuming
Solution Approach 1:
The patent segments the determination process into distinct stages: edge detection, contour extraction, shape model matching, and action pattern recognition. By dividing the complex task into manageable segments, the system maintains adaptability for various action types while reducing overall process complexity through modular processing
Solution Approach 2:
The patent applies preliminary actions by pre-processing images to extract edges and contours before action pattern recognition. Shape models are pre-established for different action categories, allowing the system to quickly match extracted contours against predefined models rather than analyzing raw images directly, thus reducing determination complexity
2Adaptability or versatility
If shape models are applied to images with various edge shapes to detect action patterns, then action categorization can be performed, but determination time increases
Solution Approach 1:
The determination process is segmented into parallel processing stages where edge detection, contour extraction, and shape model matching occur in sequence rather than requiring exhaustive analysis of all possible action patterns simultaneously, reducing overall determination time while maintaining versatility
Solution Approach 2:
Pre-established shape models for different action categories enable quick matching against extracted contours. The system performs preliminary preprocessing to extract key features (edges and contours) before pattern recognition, avoiding the need to analyze entire images and significantly reducing determination time
3Ease of operation
If conventional techniques are used to determine subject movement in captured images, then action detection can be performed, but the process requires complex algorithms
Solution Approach 1:
The patent introduces intermediary components (edge detectors, contour extractors, and shape model databases) that mediate between raw image data and action pattern recognition. These intermediaries simplify the algorithmic complexity by transforming complex image analysis into sequential, manageable operations with clear intermediate results
Data Source
AI summary
A determining method includes acquiring first direction information, when a first captured image captured by an imaging device is acquired, by referring to a storage storing a plurality of pieces of direction information indicating a plurality of directions associating with respective shapes of a plurality of contours of an object according to a plurality of directions of the object, the first direction information associated with a shape of a contour that corresponds to a shape of a contour of the subject included in the acquired first captured image among the shapes of the contours, and acquiring second direction information, when a second captured image newly captured by the imaging device is acquired, by referring to the storage to acquire second direction information associated with a shape of a contour that corresponds to a contour of the subject included in acquired second captured image among the shapes of contours.