Image Selection Using Relative Pose Features for Hidden Body Parts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current surveillance systems face challenges in accurately classifying and selecting images based on pose information, as they often rely on specific search conditions and struggle with recognizing desired states of individuals, especially when body parts are hidden or unknown.
Innovation Solution
An image selection apparatus and method that generates pose information and other relevant data from subject images, using skeleton estimation techniques to classify and select images based on feature values, allowing for flexible recognition and search of desired states without relying on specific orientations or body shapes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pose information is used for image classification and selection, then image selection accuracy is improved, but the system fails to recognize desired states when body parts are hidden or orientations are unknown
Solution Approach 1:
The patent transforms pose information from absolute coordinate values to relative feature values (distances, ratios, angles between body parts). This parameter transformation makes the representation invariant to translation, scale, and rotation, enabling accurate recognition regardless of body orientation or hidden parts while maintaining high selection accuracy
Solution Approach 2:
The patent combines multiple evaluation items (pose information, face information, body shape information, image metadata) into a composite evaluation score. This composite approach allows the system to compensate for missing or hidden body parts by weighing other available information, thereby maintaining recognition flexibility and accuracy simultaneously
2Measurement precision
If specific search conditions are applied for image classification, then classification precision is improved, but the system becomes less adaptable to unknown or hidden body states
Solution Approach 1:
The patent creates a universal pose representation system using relative feature values that can handle multiple scenarios (visible body parts, hidden body parts, different orientations, varying distances from camera) with a single unified approach. This eliminates the need for multiple specific search conditions while maintaining classification precision
Solution Approach 2:
The patent allows classification to proceed with partial pose information by using relative features that can be computed from available body parts only. The system generates evaluation scores based on the information that is present, rather than requiring complete pose data, thereby reducing complexity requirements while maintaining precision
Data Source
AI summary
A search unit of an image processing apparatus includes an information generation unit and an image selection unit. The information generation unit generates, from each of a plurality of subject images, pose information about a person included in the subject image and other information about the person. The image selection unit classifies the plurality of subject images or selects at least one target image from the plurality of subject images, by using the pose information and the other information. The search unit may further includes a query acquisition unit. The query acquisition unit acquires a query image. Then, the information generation unit further generates the pose information about the person included in the query image and the other information. The image selection unit selects at least one target image by using the pose information and the other information of the query image and each of the plurality of subject images.