Adaptive Face Depth Image Generation via Dynamic Resolution Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional 3D reconstruction technologies, particularly those using monocular speckle structured light systems, face challenges with speed and accuracy, leading to inefficient depth map generation, especially when dealing with varying face sizes, which affects real-time applications like face liveness detection.
Innovation Solution
An adaptive face depth image generation method utilizing a processor and structured light projector to downscale structured light images based on face area size, ensuring a sufficient number of depth pixels while reducing calculation complexity, allowing for consistent and efficient depth map creation across different face resolutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If depth map calculations are performed on large face areas (0.5 Megapixels), then sufficient depth information is obtained for liveness detection, but calculation time increases to hundreds of milliseconds
Solution Approach 1:
The patent implements dynamic resolution adjustment by detecting face area size and adaptively selecting appropriate depth map resolutions. For large face areas, lower resolutions (e.g., 100×100 or 150×150 pixels) are selected, while for small face areas, higher resolutions are used. This dynamic adaptation resolves the contradiction by matching resolution to actual needs, ensuring sufficient depth information for liveness detection without unnecessary computational overhead that would cause hundreds of milliseconds calculation time.
Solution Approach 2:
The patent changes the parameter of depth map resolution based on face area size detection. By establishing a relationship between face area and optimal resolution, the system transforms the fixed high-resolution approach into a variable parameter system. This resolves the technical contradiction by adjusting the resolution parameter to be sufficient but not excessive, thereby reducing calculation time while maintaining adequate depth information quality for liveness detection.
2Productivity
If depth map calculations are performed on small face areas, then calculation time is reduced to a few milliseconds, but depth information becomes insufficient for accurate liveness detection
Solution Approach 1:
The system dynamically adjusts depth map resolution based on detected face area size. When small face areas are detected, the system increases the resolution parameter to ensure sufficient depth information is captured, preventing the quality degradation that would occur with fixed low-resolution processing. This dynamic parameter adjustment resolves the contradiction by ensuring adequate depth information quality even when processing small face areas quickly.
Solution Approach 2:
The patent implements parameter changes by adjusting depth map resolution according to face area measurements. For small face areas, higher resolutions are selected to maintain sufficient depth information quality, while still benefiting from the reduced calculation time inherent in processing smaller image dimensions. This resolves the technical contradiction by optimizing the resolution parameter for each specific case.
3Device complexity
If consistent depth map resolution is used across all face sizes, then liveness algorithm training becomes simpler, but unnecessary calculations increase for large face areas
Solution Approach 1:
The patent implements a dynamic resolution selection mechanism that adjusts depth map resolution based on face area size detection. This resolves the contradiction by making the system adaptive rather than fixed: large face areas use lower resolutions to reduce unnecessary computations and energy consumption, while small face areas use higher resolutions to maintain quality. The algorithm training complexity is managed through standardized processing pipelines that can handle variable input resolutions.
4Measurement precision
If high resolution depth maps are generated for all cases, then maximum depth information is available, but real-time processing performance deteriorates
Solution Approach 1:
The patent implements dynamic resolution adjustment based on face area detection, resolving the contradiction between maximum depth information availability and real-time processing speed. The system determines the minimum necessary resolution for each specific case based on face size, ensuring adequate depth information quality while avoiding the computational overhead of consistently generating high-resolution depth maps for all cases, thereby maintaining real-time processing performance.
Solution Approach 2:
The patent changes the resolution parameter adaptively based on face area measurements. By establishing a relationship between face size and optimal resolution, the system adjusts the depth map resolution parameter to be just sufficient for each case rather than always maximum. This resolves the technical contradiction by optimizing the balance between depth information quality and processing speed through parameter adaptation.
Data Source
AI summary
An apparatus comprising an interface, a light projector and a processor. The interface may be configured to receive pixel data. The light projector may be configured to generate a structured light pattern. The processor may be configured to process the pixel data arranged as video frames comprising the structured light pattern, perform computer vision operations to detect a size of a face area of the video frames, determine a scale ratio in response to the size of the face area, extract the structured light pattern from the video frames, generate a downscaled structured light image and generate a depth map in response to the downscaled structured light image and a downscaled reference image. A downscale operation may be performed in response to the scale ratio to generate the downscaled structured light image. The scale ratio may enable the generation of the downscaled structured light image with sufficient depth pixels.