Body Part Detection Using Multi-User Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computer vision technologies face challenges in accurately detecting body parts from overlapping user images due to limitations in image division accuracy, leading to partial damage or errors in detection.
Innovation Solution
An apparatus and method that utilize a classifier trained on both single-user and multi-user image samples, employing a multi-scale ordinal pattern feature to detect body parts by dividing user images into single-user and multi-user images, and using a subwindow search to enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If user images are divided into single-user images, then detection accuracy improves, but image division accuracy becomes a limiting factor causing partial damage or errors
Solution Approach 1:
The patent divides the detection task into two separate detection systems: one for single-user images and another for multi-user images. This segmentation allows each detector to be optimized for its specific input type, avoiding the need to accurately divide overlapping user images while still achieving high detection accuracy through specialized classifiers for each scenario
Solution Approach 2:
Instead of dividing overlapping user images into separate single-user images and then detecting body parts (which causes accuracy loss), the patent inverts the approach by detecting body parts directly from the original overlapping multi-user images using a specialized multi-user detector, thereby preserving image integrity and avoiding division-related errors
2Device complexity
If a single classifier is used for all user images, then device complexity is reduced, but detection accuracy deteriorates for overlapping users
Solution Approach 1:
The patent segments the classifier system into two distinct classifiers: a single-user image classifier and a multi-user image classifier. Each classifier is trained specifically on its corresponding image type, allowing the system to maintain lower individual classifier complexity while achieving high overall detection accuracy through specialized processing for different user scenarios
Solution Approach 2:
The patent applies local quality by making different parts of the detection system have different properties: the single-user detector is optimized for non-overlapping images while the multi-user detector is optimized for overlapping images. This localized optimization ensures each detector has the appropriate complexity and characteristics for its specific function, improving overall accuracy without unnecessarily increasing total system complexity
Data Source
AI summary
An apparatus for detecting a body part from a user image may include an image acquirer to acquire a depth image, an extractor to extract the user image from a foreground of the acquired depth image, and a body part detector to detect the body part from the user image, using a classifier trained based on at least one of a single-user image sample and a multi-user image sample. The single-user image may be an image representing non-overlapping users, and the multi-user image may be an image representing overlapping users.