Human Head Detection in Depth Images Using Template Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human head detection methods, especially those based on depth information, face challenges when the subject is not facing the camera, has varying illumination, or when hands are closed or the head is occluded, and require detailed 3D human pose images that are difficult to acquire.
Innovation Solution
A head detection system that processes depth images using a head detection module with a head template, applying template matching across a depth image to identify the human head by segmenting foreground and background pixels and utilizing three concentric regions with different depth values to determine the head's location and radius.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If color-based methods or face detector based methods are used for human head location, then detection can be performed using color or grayscale intensity information, but these methods fail when the subject is not facing the camera or when illumination varies
Solution Approach 1:
The patent replaces color-based and intensity-based detection methods with depth-based detection. Instead of relying on optical properties (color, grayscale intensity) that are sensitive to illumination and pose, the invention uses depth information from depth images which provides geometric structure that is invariant to lighting conditions and subject orientation.
Solution Approach 2:
The patent changes the detection parameter from color/intensity to depth. By using depth values that represent calibrated distance in the scene, the system achieves detection that is independent of illumination variations and subject pose, directly addressing the reliability issues of conventional methods.
2Ease of manufacture
If depth-based techniques are used for human body part detection, then detection may be performed without color information, but these techniques require detailed 3D human pose images that are difficult to acquire for training
Solution Approach 1:
The patent uses a template matching approach where a predefined head template is copied and applied across the depth image. This template represents the expected depth structure of a human head, and by matching it against the depth image, the system achieves accurate detection without requiring complex 3D pose training images.
Solution Approach 2:
The patent performs preliminary action by pre-defining the head template with characteristic depth values before the actual detection process. This template is prepared in advance and can be directly applied to depth images, eliminating the need for time-consuming training with detailed 3D human pose images while maintaining detection accuracy.
3Reliability
If conventional depth-based techniques are used, then detection may be performed using only depth information, but detection becomes problematic when the subject's head is partially or fully occluded by another body part
Solution Approach 1:
The patent applies local quality by using three concentric regions with different depth values in the head template. The inner region represents the head center, the middle region represents the head periphery, and the outer region represents areas adjacent to the head. This multi-region approach allows the template matching to be more robust to occlusion by evaluating depth patterns at multiple spatial scales and locations.
Data Source
AI summary
Systems, devices and methods are described including receiving a depth image and applying a template to pixels of the depth image to determine a location of a human head in the depth image. The template includes a circular shaped region and a first annular shaped region surrounding the circular shaped region. The circular shaped region specifies a first range of depth values. The first annular shaped region specifies a second range of depth values that are larger than depth values of the first range of depth values.