Touch Classification Using Non-Bimodal ML Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current touch classification systems face challenges in distinguishing between intentional and unintentional touches, particularly when a user's palm or other body parts inadvertently contact the touch-sensitive surface, leading to incorrect interpretations and potentially unwanted device behavior.
Innovation Solution
The implementation of machine learning classifiers that differentiate between intentional and unintentional touches by analyzing frame data from touch sensors, using techniques such as blob definition, feature computation, and non-bimodal scoring to provide confidence levels in classifications, allowing for real-time differentiation without introducing latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning classifiers are implemented to differentiate between intentional and unintentional touches, then measurement precision of touch input is improved, but device complexity increases
Solution Approach 1:
The touch classification system is segmented into multiple independent machine learning classifiers, each specialized for detecting specific touch types (e.g., intentional touches, unintentional touches, palm touches, finger touches). This segmentation allows each classifier to focus on specific patterns, improving overall classification precision while maintaining manageable complexity through modular design.
Solution Approach 2:
The patent introduces machine learning classifiers as intermediary components between the touch sensor and the user interface processing system. These classifiers act as intelligent mediators that analyze touch patterns and provide classified touch type information to downstream systems, enabling accurate differentiation between intentional and unintentional touches without requiring complex processing throughout the entire system.
2Reliability
If machine learning classification is applied to all touch events, then reliability of touch recognition is improved, but processing time increases
Solution Approach 1:
The system applies machine learning classification selectively rather than uniformly to all touch events. Classifiers are applied based on specific conditions and touch patterns that require differentiation, allowing the system to maintain high reliability for critical classifications while avoiding unnecessary processing overhead for straightforward touch events.
Solution Approach 2:
The patent replaces traditional rule-based or threshold-based touch classification mechanisms with machine learning classifiers. This substitution enables more accurate and reliable classification of complex touch patterns while the classifiers are optimized to process data efficiently, reducing the time penalty compared to simpler but less reliable methods.
3Measurement precision
If multiple machine learning classifiers are used to differentiate touch types, then measurement precision is improved, but use of energy increases
Solution Approach 1:
The use of multiple specialized machine learning classifiers is segmented and organized efficiently, with each classifier handling specific touch type differentiation tasks. This segmentation allows the system to achieve high measurement precision for various touch types while managing energy consumption through targeted rather than exhaustive classification approaches.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A method for touch classification includes obtaining frame data representative of a plurality of frames captured by a touch-sensitive device, analyzing the frame data to define a respective blob in each frame of the plurality of frames, the blobs being indicative of a touch event, computing a plurality of feature sets for the touch event, each feature set specifying properties of the respective blob in each frame of the plurality of frames, and determining a type of the touch event via machine learning classification configured to provide multiple non-bimodal classification scores based on the plurality of feature sets for the plurality of frames, each non-bimodal classification score being indicative of an ambiguity level in the machine learning classification.