Human Body Detection via Head Estimation and Overlap Ratios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human body detection techniques face challenges in accurately detecting human bodies in scenes where objects or people overlap, leading to shielding issues, which result in incomplete information and incorrect bounding box estimation, especially in crowded environments.
Innovation Solution
A human body detection device and method that estimates specific parts of the body, such as the head, to predict the existence of the human body candidate area and subsequently the human body rectangle, while considering overlap ratios between partial and human body rectangles to prevent incorrect integration of different persons as a single entity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If NMS processing is used to integrate multiple bounding boxes for the same object, then the number of detected objects is reduced, but correct detections may be rejected as false positives
Solution Approach 1:
The patent segments the object detection process into two distinct stages: part detection (detecting specific body parts like heads) and whole object detection (detecting complete human bodies). This segmentation allows the system to use different integration strategies for each stage, applying NMS selectively to part detections while preserving whole object detections, thereby reducing false rejections of correct detections.
Solution Approach 2:
The patent introduces an intermediary relationship between part detections and whole object detections. Instead of directly integrating all bounding boxes uniformly, the system uses part detections as intermediate information to guide and refine whole object detection. This intermediary layer helps distinguish between actual overlaps (different people) and detection variations (same person), preventing correct detections from being rejected.
2Productivity
If detection is performed in crowded environments with overlapping people, then more people can be monitored, but detection accuracy decreases due to shielding
Solution Approach 1:
The patent divides the human body into detectable parts (head, torso, limbs) and detects these parts separately before integrating them into complete body detections. In crowded environments, even when bodies are shielded or overlapping, certain body parts remain visible and can be detected. This segmentation enables the system to maintain monitoring capacity in crowded scenes while improving detection accuracy by reconstructing complete bodies from visible parts.
Solution Approach 2:
The patent transitions from detecting only complete human bodies (2D bounding boxes) to detecting both body parts and whole bodies across multiple detection levels. By adding the dimension of part-level detection, the system can infer the presence and position of complete bodies even when partially occluded, thereby maintaining productivity in crowded environments while improving measurement precision.
3Device complexity
If only whole body detection is performed, then the detection process is simple, but detection fails when the body is shielded or occluded
Solution Approach 1:
The patent segments the detection task into part detection and whole object detection components. Instead of attempting to detect complete bodies in a single step, the system first detects prominent body parts (which are more likely to be visible even when shielded) and then uses these part detections to infer and confirm whole body positions. This segmentation significantly improves detection robustness in occluded scenarios while adding only moderate complexity through the two-stage process.
Solution Approach 2:
The patent performs preliminary detection of body parts before performing whole object detection. By first identifying visible body parts (such as heads or torsos) that are less likely to be completely occluded, the system establishes preliminary detection results that guide the subsequent whole body detection process. This preliminary action improves reliability by providing anchor points for detecting shielded bodies, while the modular structure keeps overall complexity manageable.
Data Source
AI summary
The human body detection device estimates a head from an image, and outputs coordinates and a confidence level of the head rectangle including the head. Next, the human body detection device estimates and outputs a human body candidate area in which the human body corresponding to the head is predicted to exist based on the coordinates of the head rectangle. Then, the human body detection device estimates the human body rectangle including the human body based on the human body candidate area, and outputs coordinates and a confidence level of the human body rectangle.