Real-time Face Detection via Parallel Multi-scale Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face detection methods struggle to process images in real-time due to the need for sequential scaling and analysis of faces with varying sizes, which complicates the detection process and reduces efficiency.
Innovation Solution
A real-time face detection apparatus that down-scales input images at multiple ratios and applies classifiers in parallel to determine face regions, using a look-up table for pixel-based confidence values to identify faces across different scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sequential scaling and face detection is performed on images with various sizes, then detection accuracy for different face sizes is improved, but processing time increases and real-time performance deteriorates
Solution Approach 1:
The image processing is segmented into multiple scale versions (e.g., original scale and down-scaled versions). Each scale is processed independently by dedicated hardware units in parallel, eliminating the need for sequential processing while maintaining detection accuracy across different face sizes.
Solution Approach 2:
The solution transitions from sequential processing in time dimension to parallel processing by adding spatial dimension through multiple hardware units. Different scale images are processed simultaneously by separate hardware components, achieving real-time performance without sacrificing detection accuracy.
2Reliability
If multiple classifiers are applied to detect faces of various sizes, then detection performance is improved, but device complexity and hardware resources increase
Solution Approach 1:
Multiple face detection hardware units are designed with identical classifier structures that can process different scale images. Each unit is universal and can handle any scale by receiving appropriately scaled input images, reducing overall system complexity compared to having specialized classifiers for each scale.
Solution Approach 2:
The classifier is copied into multiple identical hardware units, each processing a specific scale. This copying approach allows parallel processing while maintaining simplicity, as each copy is a standalone unit with the same structure, avoiding the need for complex adaptive systems.
3Adaptability or versatility
If images are scaled to various sizes for accurate face detection, then detection capability for different face sizes is improved, but processing speed and real-time capability deteriorate
Solution Approach 1:
Image scaling is performed as a preliminary action before detection, with dedicated hardware units preparing down-scaled versions of the input image. This pre-processing enables the main detection units to operate in parallel without waiting for sequential scaling, maintaining both adaptability and speed.
Solution Approach 2:
The patent replaces software-based sequential processing with hardware-based parallel processing. Dedicated hardware units perform scaling and detection simultaneously, substituting the mechanical sequential operations with concurrent hardware operations, thereby achieving real-time performance.
Data Source
AI summary
Disclosed herein is a real-time face detection apparatus. The real-time face detection apparatus includes a down-scaling unit and a face region comparison unit. The down-scaling unit down-scales an input image at at least one ratio. The face region comparison unit creates a plurality of windows for the image down-scaled at the at least one ratio, acquires face region confidence of each of window images within the created windows by comparing the window image with a classifier, and determines whether the window image corresponds to a face region.