Eye-Tracking Feature Extraction via Sub-Frame Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing eye-tracking technologies consume high power and experience delays due to the need to process complete eye frames, which is inefficient and requires significant hardware resources.
Innovation Solution
The method involves dividing eye frames into sub-frames and using multiple feature extraction stages, where launch features are extracted from sub-frames sequentially, allowing for reduced storage and processing requirements, and terminal features are computed using superposition operations to determine the eye's gazing direction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complete eye frames are processed to compute gazing direction, then measurement precision is improved, but power consumption increases and computation delay occurs
Solution Approach 1:
The patent divides the complete eye frame into multiple sub-frames and processes them through multiple feature extraction stages. Instead of loading and processing the entire eye frame at once, the system extracts features from individual sub-frames sequentially, reducing the amount of data that needs to be stored and processed simultaneously, thereby lowering power consumption while maintaining gazing direction accuracy.
Solution Approach 2:
The patent performs preliminary feature extraction from sub-frames before computing the final gazing direction. By extracting intermediate features from each sub-frame in advance and storing only these extracted features rather than the complete raw eye frame data, the system reduces subsequent computation requirements and power consumption while preserving the information needed for accurate gazing direction calculation.
2Measurement precision
If complete eye frames are stored for processing, then measurement precision is improved, but hardware area increases
Solution Approach 1:
The patent segments the eye frame data into multiple sub-frames and processes them through multiple extraction stages. The hardware only needs to store and process a portion of the eye frame at any given time rather than the complete frame, significantly reducing the memory and processing unit area required while maintaining the capability to compute accurate gazing direction through sequential processing of all sub-frames.
Solution Approach 2:
The patent extracts only the necessary features from sub-frames rather than storing and processing the complete eye frame data. By taking out and storing only the extracted features from each sub-frame in the buffer memory, the system reduces the hardware area required for data storage and processing while preserving the essential information needed for gazing direction computation.
3Measurement precision
If complete eye frames are processed, then measurement precision is improved, but computation delay increases
Solution Approach 1:
The patent divides the eye frame processing into multiple sub-frames and multiple feature extraction stages that can be executed sequentially and potentially in parallel. This segmentation allows the system to start processing earlier and spread the computation over time, reducing the peak computation delay while maintaining accurate gazing direction measurement through the aggregation of features from all sub-frames.
Solution Approach 2:
The patent performs preliminary feature extraction from sub-frames and stores the extracted features in buffer memory before the final gazing direction computation is needed. This preliminary action allows the system to prepare data in advance, reducing the computation delay at the critical moment when gazing direction information is required, while ensuring measurement precision is maintained through complete feature extraction.
Data Source
AI summary
An image operation method and system for obtaining an eye's gazing direction are provided. The method and system employ multiple extraction stages for extracting eye-tracking features. An eye frame is divided into sub-frames, which are then sequentially temporarily stored in a storage unit. Launch features of sub frames are sequentially extracted from the sub frames by a first feature extraction stage, where a data of a former sub-frame is extracted before a data of a latter sub-frame is needed to be stored. Next, the remaining feature extraction stages apply a superposition operation on the launch features to obtain terminal features, which are then computed to obtain an eye's gazing direction.