In-Vehicle Camera Driver Face Detection Under Non-Uniform Illumination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for detecting facial features in vehicles struggle with non-uniform illumination changes, leading to inaccurate detection due to local variations in light exposure, which complicates the identification of facial contours and false detection of bright areas as facial features.
Innovation Solution
An image processing method that integrates pixel brightness in one direction and analyzes changes in the other direction to identify candidate locations, combines brightness of adjoining pixels with subtraction of pixels at a distance, and uses mean and variance values to select the most effective detection method based on illumination conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If automatic gain function is used to adjust brightness of the entire image, then the brightness becomes constant under uniform illumination, but false detection occurs under non-uniform illumination (e.g., only bright parts are detected as facial contour)
Solution Approach 1:
The patent segments the image processing into multiple detection methods: (1) facial contour detection based on brightness gradients, (2) eye detection based on dark region analysis, and (3) geometric relationship verification. This segmentation allows each method to focus on specific features, reducing false detections under non-uniform illumination while maintaining reliability.
Solution Approach 2:
Instead of relying on a single automatic gain function for the entire image, the patent applies partial actions: local brightness adjustment for contour detection, separate thresholding for eye detection, and selective verification of geometric relationships. This partial approach prevents over-correction that causes false detections while maintaining detection accuracy.
2Measurement precision
If multiple detection methods are used to handle non-uniform illumination, then detection accuracy improves, but device complexity increases
Solution Approach 1:
The patent implements dynamic selection and combination of detection methods based on image characteristics. The system adaptively switches between facial contour detection, eye detection, and geometric verification depending on illumination conditions, optimizing detection accuracy while managing processing complexity through conditional execution.
Solution Approach 2:
The patent merges multiple detection methods into a unified processing framework where facial contour detection, eye detection, and geometric relationship verification work together. The results from different methods are combined and cross-validated, improving measurement precision while the integrated structure manages complexity through coordinated processing.
Data Source
AI summary
Provided are an image processing method, an image processing system, an image processing device and a computer program for detecting a detection object such as nares of the driver with a high degree of accuracy in, for example, a system using an in-vehicle camera which is mounted on a vehicle and takes an image of the face of the driver. A detection object is diversified by a variety of detection methods such as a method for detecting a plurality of locations in the vertical direction as candidates during image pickup of an image, detecting a range to be a candidate of a detection object on the basis of the brightness of a pixel for each of rows of pixels lined up in the horizontal direction corresponding to each detected location and specifying a detection object from candidates of a detection object on the basis of the length of the detected range.