Vehicle Gaze Estimation Using Candidate-Specific Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for detecting the gaze of an occupant in a mobile object fail to accurately consider the type of gaze candidate towards which the occupant is directing their gaze.
Innovation Solution
A gaze estimation device that calculates probability values for each gaze candidate based on the occupant's gaze and probability distributions specific to each type of gaze candidate, determining the likelihood of the occupant directing their gaze towards a particular candidate.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If gaze detection is performed without considering gaze candidate types, then the detection process is simple, but the detection accuracy is low
Solution Approach 1:
The patent applies parameter changes by introducing different probability distribution parameters specific to each gaze candidate type (dashboard, mirror, external object). Each gaze candidate type has its own probability distribution characteristics, allowing the system to accurately model and detect gaze direction by selecting appropriate parameters based on the candidate type, thereby improving detection accuracy without significantly increasing system complexity
Solution Approach 2:
The patent implements local quality by assigning different probability distribution characteristics to different gaze candidate types. Each gaze candidate (dashboard, mirror, external object) has locally optimized probability distribution parameters that reflect the specific viewing characteristics of that region, enabling accurate gaze detection tailored to each candidate type while maintaining overall system efficiency
2Measurement precision
If multiple probability distributions are used for different gaze candidate types, then the gaze detection accuracy improves, but the calculation complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the gaze detection problem into distinct segments corresponding to different gaze candidate types (dashboard, mirror, external object). Each segment has its own probability distribution model, allowing the system to process and calculate probabilities separately for each type, which improves accuracy while managing calculation complexity through modular processing
Solution Approach 2:
The patent uses parameter changes by maintaining different probability distribution parameters for each gaze candidate type. The system dynamically selects and applies appropriate parameters based on the detected gaze candidate type, enabling accurate multi-type gaze detection without requiring a single complex unified model, thus balancing accuracy and computational efficiency
Data Source
AI summary
Provided is a gaze estimation device for estimating toward which gaze candidate an occupant of a mobile object is directing his or her gaze from among a plurality of gaze candidates including gaze candidates existing on a structure of the mobile object and an object near the mobile object, the gaze estimation device being configured to: calculate, for each of the plurality of gaze candidates, a probability value of the occupant directing his or her gaze toward the gaze candidate based on the gaze and a probability distribution representing probabilities of directing his or her gaze toward each of the plurality of gaze candidates; and estimate that the occupant is directing his or her gaze toward a gaze candidate for which the probability value is equal to or larger than a threshold value among the plurality of gaze candidates.