Human Segmentation Using Depth Data and Face Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing technologies face difficulties in accurately separating human body regions from backgrounds, especially when the first frame contains humans, and often misclassify moving objects other than humans as human bodies.
Innovation Solution
An image processing apparatus and method that utilize depth data and face detection technology to accurately isolate human body regions by comparing depth data with background models, employing hard and soft constraint conditions for tracking and expansion of human body regions, and refining the extraction process using post-processing techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional background modeling is used to extract human body regions, then moving regions can be separated, but humans in the first frame cannot be detected and other moving objects are misclassified as humans
Solution Approach 1:
The patent segments the detection process into multiple specialized units: face detecting unit for detecting human faces, human body region extracting unit for extracting human body regions, and other object region extracting unit for extracting other moving objects. This segmentation allows each unit to focus on specific detection tasks, improving overall accuracy in distinguishing humans from other objects while reliably detecting humans in the first frame.
Solution Approach 2:
The patent introduces depth data as an intermediary element to bridge color image data and human body region extraction. The depth data providing unit supplies depth information that serves as a mediator to verify whether detected regions correspond to human bodies, enabling accurate differentiation between humans and other moving objects while detecting stationary humans in the first frame.
2Productivity
If all moving regions are extracted as human body regions, then detection speed is maintained, but accuracy decreases due to misclassification of other objects
Solution Approach 1:
The patent segments the extraction process into parallel specialized units that operate simultaneously: face detecting unit, human body region extracting unit, and other object region extracting unit. This segmentation enables efficient parallel processing that maintains detection speed while improving accuracy by directing each unit to perform its specific extraction function without misclassification.
Solution Approach 2:
The patent changes the parameter basis for extraction by introducing depth data as an additional dimension beyond color information. The human body region extracting unit uses depth data parameters to verify human body regions, while the other object region extracting unit identifies regions with different depth characteristics, maintaining high-speed extraction with improved accuracy through multi-parameter verification.
3Measurement precision
If depth data is used for human body region extraction, then accuracy improves, but processing time increases
Solution Approach 1:
The patent segments the depth data processing into specialized extraction units that operate in parallel. The human body region extracting unit and other object region extracting unit simultaneously process depth data using different criteria, enabling efficient utilization of depth information without sequential processing delays, thus improving accuracy while minimizing processing time.
Solution Approach 2:
The patent applies partial action by having different extraction units focus on specific regions of interest identified through face detection. The human body region extracting unit concentrates processing on regions near detected faces using depth data, rather than processing the entire image, which reduces overall processing time while maintaining high accuracy in critical human body region extraction.
Data Source
AI summary
An image processing apparatus and method may accurately separate only humans among moving objects, and also accurately separate even humans who have no motion via human segmentation using a depth data and face detection technology. The apparatus includes a face detecting unit to detect a human face in an input color image, a background model producing/updating unit to produce a background model using a depth data of an input first frame and face detection results, a candidate region extracting unit to produce a candidate region as a human body region by comparing the background model with a depth data of an input second or subsequent frame, and to extract a final candidate region by removing a region containing a moving object other than a human from the candidate region, and a human body region extracting unit to extract the human body region from the candidate region.