Pupil Detection Device Using Filter Weight Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pupil detection devices face challenges in accurately detecting pupils in input images due to high rates of false detections, which affect their performance and reliability.
Innovation Solution
A pupil detection device that employs a filter weight learning unit to generate a target image and learn filter weights, a split image generation unit to create a pupil region split image, and a pupil coordinate detection unit to remove noise and detect pupil coordinates, utilizing a combination of convolution, max-pooling, and up-sampling filters, along with a skip connecting network and auxiliary filters to improve image processing accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional pupil detection methods are used, then the device structure remains simple, but false detection rate increases and detection accuracy deteriorates
Solution Approach 1:
The detection device is segmented into multiple functional units: filter weight learning unit, split image generation unit, and pupil coordinate detection unit. Each unit performs a specific function in the detection pipeline, allowing complex processing to be organized into manageable modules that work together to improve accuracy while maintaining structural clarity
Solution Approach 2:
The filter weight learning unit performs preliminary learning and weight optimization before actual pupil detection. By pre-learning optimal filter weights from training data and storing them for later use, the system prepares the detection mechanism in advance, enabling accurate detection without requiring complex real-time computations during actual operation
Solution Approach 3:
A pupil region split image is introduced as an intermediary representation between the original input image and the final pupil coordinate detection. This intermediate image highlights the pupil region specifically, making the subsequent coordinate detection more accurate while keeping the overall system structure manageable
2Measurement precision
If filter weights are learned and optimized, then detection accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
Filter weight learning is performed in advance during an offline training phase, and the learned weights are stored for reuse. During actual pupil detection operations, the pre-learned weights are applied directly without requiring real-time learning computations, significantly reducing processing time while maintaining high detection accuracy
Solution Approach 2:
The computationally intensive filter weight learning process is extracted from the real-time detection pipeline and performed separately during training. Only the lightweight application of learned weights and simple coordinate detection remains in the real-time system, separating heavy computation from time-critical operations
Data Source
AI summary
A pupil detection device includes: a filter weight learner configured to generate a target image on the basis of pupil coordinate data acquired from a learning image and learn weights of a plurality of filters in order to generate a filtered image, by filtering the learning image using the plurality of filters, that is within a predetermined reference range of the target image; a split image generator configured to generate a pupil region split image for an input image using the plurality of filters having the learned weights; and a pupil coordinate detector configured to remove noise of the pupil region split image, select at least one of a plurality of pixels from which noise is removed, and detect pupil coordinates.