Human Body Detection via Portion-Body Rectangle Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human body detection technologies face challenges in accurately detecting humans when they are shielded by obstacles or overlapping with others, leading to reduced accuracy in bounding box estimation and integration of duplicate rectangles.
Innovation Solution
The proposed solution involves an information processing apparatus and method that estimates specific portions of the human body, such as the head, and integrates duplicate human body rectangles based on the relationship between portion and human body rectangles, prioritizing rectangles derived from the human body to maintain accuracy in crowded and shielded environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human body detection is performed using conventional object detection techniques, then detection can be achieved in general conditions, but detection accuracy deteriorates when humans are shielded by obstacles or overlapping with others
Solution Approach 1:
The patent segments the human body detection task into two distinct detection processes: one detecting the entire human body and another detecting specific portions (head, upper body, lower body). By dividing the detection object into multiple parts and detecting them separately, the system can identify humans even when partially shielded, as long as any portion is visible. This segmentation approach directly resolves the contradiction by maintaining detection accuracy in shielded environments where full-body detection would fail.
Solution Approach 2:
The patent introduces a new dimension to human body detection by adding portion-level detection alongside traditional full-body detection. Instead of relying solely on one detection method, the system operates in multiple detection dimensions (whole body vs. specific portions), allowing it to cross-validate results and maintain reliability even when one dimension fails due to shielding.
2Reliability
If multiple detection methods are used to improve accuracy in shielded environments, then detection robustness is improved, but system complexity increases
Solution Approach 1:
The patent segments both the detection target (human body into portions) and the detection process (separate detection models for whole body and portions). This segmentation allows the system to use simpler, specialized detectors for each task rather than one complex detector trying to handle all scenarios, thereby managing system complexity while improving robustness.
Solution Approach 2:
The patent merges the results from multiple detection methods (full-body detection and portion detection) through an integration process. By combining the outputs of simpler individual detectors rather than using one complex detector, the system achieves improved robustness while keeping individual components manageable in complexity.
3Productivity
If duplicate human body rectangles are integrated based on traditional methods, then processing speed is maintained, but accuracy deteriorates due to improper integration of rectangles from different detection sources
Solution Approach 1:
The patent applies local quality by treating rectangles from different detection sources (full-body vs. portion-based) differently during integration. Instead of applying a uniform integration rule to all rectangles, the system identifies the source of each rectangle and applies appropriate integration logic, prioritizing full-body detection results when available and using portion-based results as supplements. This localized approach to integration maintains both speed and accuracy.
Solution Approach 2:
The patent introduces an intermediary integration process that acts as a mediator between the full-body detection results and portion-based detection results. This intermediary step intelligently combines the two sources, resolving conflicts and merging complementary information, thereby maintaining integration speed while improving the accuracy of the final bounding box estimates.
Data Source
AI summary
A portion rectangle estimation unit estimates a specific portion of a human body from an image, and outputs a first portion rectangle. A first human body rectangle estimation unit estimates the human body corresponding to the specific portion in the first portion rectangle based on coordinates of the first portion rectangle, and outputs a first human body rectangle. A second human body rectangle estimation unit estimates the human body from the image, and outputs a second human body rectangle. A second portion rectangle estimation unit estimates the specific portion corresponding to the human body in the second human body rectangle based on coordinates of the second human body rectangle, and outputs a second portion rectangle. A human body integration unit integrates duplicate human rectangles based on at least a relationship between the portion rectangles with respect to a plurality of pairs of mutually corresponding portion rectangles and human body rectangles.