Driver Abnormality Detection via Saliency Dispersion Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing driver abnormality detection systems face challenges in accurately determining the agreement degree between a driver's visual line direction and high-saliency regions due to high computational loads, which limits their precision with limited in-vehicle computation resources.
Innovation Solution
A driver abnormality sign detection device that uses a control circuit to detect abnormalities by acquiring visual line information and image data, calculating dispersion in saliency distributions, and applying correction values to predict saccade frequencies and amplitudes, thereby reducing the need for high-resolution saliency distribution calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution saliency distribution data are acquired to accurately specify the agreement degree between visual line direction and high-saliency region, then measurement precision is improved, but computation load increases making it difficult to calculate with in-vehicle computer resources
Solution Approach 1:
The patent extracts only the essential feature (dispersion of high-saliency section) from the complete saliency distribution, rather than processing the entire high-resolution saliency map. This extraction approach maintains detection precision while significantly reducing computation load by focusing only on the dispersion characteristic of the high-saliency region.
Solution Approach 2:
Instead of calculating the complete high-resolution saliency distribution, the patent performs partial action by calculating only the dispersion of the high-saliency section. This partial calculation approach provides sufficient information for abnormality detection without the excessive computational burden of full-resolution processing.
2Measurement precision
If high-resolution saliency distribution data are calculated to accurately determine visual line agreement with high-saliency region, then detection precision is improved, but productivity decreases due to high computation load
Solution Approach 1:
The patent extracts only the dispersion characteristic of the high-saliency section from the complete saliency distribution. This extraction enables accurate specification of visual line agreement while dramatically improving processing speed by avoiding calculation of the entire high-resolution saliency map.
Solution Approach 2:
The patent performs partial calculation by computing only the dispersion of the high-saliency section rather than the full saliency distribution. This partial action maintains sufficient detection precision while significantly enhancing processing productivity through reduced computational requirements.
3Measurement precision
If the agreement degree between visual line direction and high-saliency region is accurately specified, then detection precision is improved, but computation resources are insufficient for high-resolution saliency calculation
Solution Approach 1:
The patent extracts only the essential dispersion feature from the high-saliency section, requiring minimal computation resources while maintaining accurate abnormality detection. This extraction approach enables precision detection within the constraints of limited in-vehicle computer resources.
Solution Approach 2:
The patent performs partial calculation of only the high-saliency section dispersion rather than the complete saliency distribution. This partial action reduces resource consumption to levels compatible with in-vehicle computers while preserving sufficient detection precision.
Data Source
AI summary
A driver abnormality sign detection device includes a control circuit configured to detect an abnormality sign of a driver based on visual line information acquired from a visual line detector and on image data from a vehicle-outside camera. The control circuit is configured to acquire reference saccades of the driver, calculate dispersion of a section including a peak of saliency in a visual field of the driver based on the image data, calculate predicted saccades by correcting the reference saccades with correction values acquired based on the dispersion of the section including the peak of the saliency, and detect presence or absence of the abnormality sign of the driver based on a visual line abnormality degree representing an extent that measured saccades diverge from the predicted saccades.