Method and electronic device for accidental touch prediction using ml classification

Patent metadata

Field	Value
Publication number	US-11442579-B2
Application number	US-202017116455-A
Country	US
Kind code	B2
Filing date	Dec 9, 2020
Priority date	Dec 13, 2019
Publication date	Sep 13, 2022
Grant date	Sep 13, 2022

Abstract

A method for accidental touch prediction using machine learning (ML) classification is provided. The method includes determining, by an electronic device, a mutual data index of a sensor data using a first ML model. Further, the method includes recognizing whether the sensor data corresponds to an object touch or a non-object touch based on the mutual data index. Further, the method includes performing, by the electronic device, one of detecting that the electronic device is in a pocket mode and providing an object touch notification on a touch screen of the electronic device in response to determining that the sensor data corresponds to the object touch, and recognizing whether the sensor data corresponds to an accidental touch or a non-accidental touch using at least one second ML model based on touch data in response to determining that the sensor data corresponds to the non-object touch.

First claim

What is claimed is:

1. A method for accidental touch prediction using machine learning (ML) classification by an electronic device, the method comprising:
capturing, by the electronic device, sensor data corresponding to a touch on a touch screen of the electronic device;
determining, by the electronic device, a mutual data index of the sensor data using a first ML model;
recognizing, by the electronic device, whether the sensor data corresponds to an object touch or a non-object touch based on the mutual data index; and
performing, by the electronic device, one of:
in response to determining that the sensor data corresponds to the object touch, detecting that the electronic device is in a pocket mode and providing an object touch notification, or
in response to determining that the sensor data corresponds to the non-object touch, recognizing whether the sensor data corresponds to an accidental touch or a non-accidental touch using at least one second ML model.

2. The method of claim 1 , wherein the recognizing of whether the sensor data corresponds to the object touch or the non-object touch based on the mutual data index comprises:
extracting a relationship among mutual data indices;
determining probability information based on the relationship;
determining whether the probability information exceeds a probability criterion; and
performing one of:
in response to determining that the probability information does not exceed the probability criterion, recognizing that the sensor data corresponds to the object touch, or
in response to determining that the probability information exceeds the probability criterion, recognizing that the sensor data corresponds to the non-object touch.

3. The method of claim 1 , wherein the recognizing of whether the sensor data corresponds to the accidental touch or the non-accidental touch using the at least one second ML model comprises:
classifying the sensor data by running the at least one second ML model using one or more electronic device features with the mutual data index;
determining whether the sensor data corresponds to the accidental touch or non-accidental touch using the at least one second ML model based on the classifying;
in response to determining that the sensor data corresponds the non-accidental touch, detecting that the electronic device is in a non-pocket mode; and
in response to determining that the sensor data corresponds to the accidental touch, determining luminance information of the electronic device and detecting whether the electronic device is in the pocket mode or the non-pocket mode based on the luminance information of the electronic device.

4. The method of claim 3 , wherein the detecting of whether the electronic device is in the pocket mode or the non-pocket mode based on the luminance information of the electronic device comprises:
determining whether the luminance information meets a luminance criterion; and
performing one of:
in response to determining that the luminance information meets the luminance criterion, detecting that the electronic device is in the non- pocket mode, or
in response to determining that the luminance information does not meet the luminance criterion, detecting that the electronic device is in the pocket mode and providing an accidental touch notification.

5. The method of claim 1 ,
wherein the recognizing of whether the sensor data corresponds to the accidental touch or the non-accidental touch using the at least one second ML model comprises:
classifying the sensor data by executing the at least one second ML model, the at least one second ML model comprising at least one of a random forest neural network, an extreme gradient boosting tree neural network, a gradient boosting tree neural network, or a support vector machine neural network;
determining whether the sensor data corresponds to the accidental touch or the non-accidental touch using the at least one second ML model based on the classifying;
in response to determining that the sensor data corresponds the non-accidental touch, detecting that the electronic device is in the pocket mode; and
in response to determining that the sensor data corresponds to the accidental touch, determining probability information associated with the at least one second ML model and detecting whether the electronic device is in the pocket mode or a non-pocket mode based on the probability information, and

wherein the probability information is obtained based on a weightage factor comprising at least one of a training error value associated with the at least one second ML model, a mean runtime value of a validation set associated with the at least one second ML model, or an update time of a model parameter associated with the at least one second ML model.

6. The method of claim 5 , wherein the detecting of whether the electronic device is in the pocket mode or the non-pocket mode based on the probability information comprises:
determining whether the probability information meets a probability criterion; and
performing one of:
in response to determining that the probability information meets the probability criterion, detecting that the electronic device is in the non-pocket mode, or
in response to determining that the probability information does not meet the probability criterion, detecting that the electronic device is in the pocket mode and providing an accidental touch notification.

7. The method of claim 1 , wherein the mutual data index indicates a resistance of a conductive object or a finger of a user that comes in contact with the touch screen of the electronic device.

8. The method of claim 1 , wherein the first ML model is created and trained by:
obtaining a plurality of mutual data indices from a plurality of users and objects, each mutual data index indicating the sensor data corresponding to electronic devices of a user and objects;
extracting local special features from the mutual data index of each user using a kernel operation, the local special features being extracted based on a resistivity of a finger or a resistivity of an object that comes in contact with the touch screen of the electronic device;
generating a heat map and a probability of an abnormal touch based on the local special features; and
creating and training the first ML model using the heat map and the probability of the abnormal touch.

9. The method of claim 1 , wherein the at least one second ML model is created and trained by:
receiving a plurality of feature datasets;
determining an optimal feature that changes an impurity of each column and each value in respective columns of electrodes associated with each feature dataset;
generating at least one rule to split each feature dataset based on a maximum depth associated with the optimal feature;
generating a plurality of classifiers based on the at least one rule; and
creating and training the at least one second ML based on the plurality of classifiers.

10. The method of claim 1 , wherein the at least one second ML model is created and trained by:
obtaining data of a plurality of predefined accidental touches and data of a plurality of predefined non-accidental touches;
obtaining a plurality of predefined electronic device features for the data of the plurality of predefined accidental touches and the data of the plurality of predefined non-accidental touches;
building at least one classifier for the plurality of predefined electronic device features;
simultaneously training the at least one classifier for the plurality of predefined electronic device features; and
creating and training the at least one second ML model based on the at least one trained classifier.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Agarwal Ankit
• Rathour Sunil
• Oberoi Harshit
• Goyal Nitesh
• Tanwar Nitin Kishorbhai
• Khushu Vipin
• Choudhary Choice

Classifications

• G06F3/0418 Primary

  for error correction or compensation, e.g. based on parallax, calibration or alignment · CPC title

• G06F3/04186 Primary

  Touch location disambiguation · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06F18/24

  Classification techniques · CPC title