Curvature-Based Face Detection in Depth Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for processing depth maps struggle to accurately identify and extract humanoid faces from scenes, particularly in cases where the head is tilted or partially obscured, due to limitations in curvature analysis and filter sensitivity to pitch and yaw angles.
Innovation Solution
The method involves extracting a curvature map from depth maps, detecting blobs with convex surfaces, and applying curvature filters to correct for roll angles and scale, using specialized filter kernels and log likelihood calculations to confidently identify face locations, even at varying resolutions and orientations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional curvature analysis methods are used for face detection, then the detection process is simple, but the accuracy deteriorates when heads are tilted or partially obscured
Solution Approach 1:
The patent transforms the depth map into a curvature map by computing curvature values at each pixel location. This parameter transformation allows the detection method to identify facial features through curvature patterns rather than raw depth values, improving accuracy for tilted or partially obscured heads while maintaining computational feasibility through efficient curvature calculation algorithms
Solution Approach 2:
The patent introduces a new dimensional representation by creating a curvature map from the depth map. Instead of analyzing facial features directly in the depth domain, the method operates in the curvature domain where facial structures exhibit distinctive patterns, thereby improving detection robustness without proportionally increasing complexity
2Adaptability or versatility
If pitch and yaw sensitive filters are used, then the filter can detect specific orientations, but the versatility deteriorates due to sensitivity to head orientation variations
Solution Approach 1:
The patent inverts the traditional approach by making the filter orientation-insensitive rather than orientation-specific. Instead of designing filters that respond to particular pitch and yaw angles, the method creates filters that detect curvature patterns invariant to head orientation, thereby achieving versatility across different facial poses while maintaining detection precision through the inherent orientation-selectivity of curvature features
Solution Approach 2:
The curvature-based filters serve multiple functions simultaneously: they detect facial features across various orientations, distinguish between different facial structures, and work with partially obscured faces. This multi-functionality is achieved by designing filters that respond to fundamental curvature patterns rather than orientation-specific features, enabling a single filter set to handle diverse facial configurations
Data Source
AI summary
A method for processing data includes receiving a depth map of a scene containing at least a humanoid head, the depth map comprising a matrix of pixels having respective pixel depth values. A digital processor extracts from the depth map a curvature map of the scene. The curvature map includes respective curvature values of at least some of the pixels in the matrix. The curvature values are processed in order to identify a face in the scene.