Probabilistic palm rejection using spatiotemporal touch features and iterative classification

• US 10,031,619 B2
• Filed: 04/14/2015
• Issued: 07/24/2018
• Est. Priority Date: 04/14/2014
• Status: Active Grant

• Current Assignee: Carnegie Mellon University
• Original Assignee: Carnegie Mellon University
• Inventors: Schwarz, Julia, Harrison, Christopher
• Primary Examiner(s): Yang, Nan-Ying
• Application Number:
  US15/125,663
• Publication Number:
  US 20170024073A1
• Time in Patent Office: 1,197 Days
• CPC Class Codes: G06F 3/04186 Touch location disambiguation

Abstract

The present invention is a palm rejection technique utilizing temporal features, iterative classification, and probabilistic voting. Touch events are classified based on features periodically extracted from time windows of increasing size, always centered at the birth of the event. The classification process uses a series of decision trees acting on said features.

Claims

1. A method for distinguishing between a palm touch and a pointer touch on a touch-sensitive screen of a computing device, comprising:

receiving notification of a touch event associated with the touch-sensitive screen from the computing device;

classifying the touch event based on a series of features that characterize the touch event and a relationship of the touch event to other events; and

periodically re-classifying the touch event over a time window centered at a time the touch event was received, wherein, for a re-classification of the re-classifying of the touch event, the time window expands symmetrically an equal amount of time forward and backward from the time the touch event was received, while maintaining the time window centered at the time, wherein, to facilitate the re-classifying of the touch event, the re-classification is based on the series of features, previous features obtained from a previous touch event associated with the touch-sensitive screen that occurred prior to the time, and subsequent features obtained from a subsequent touch event associated with the touch-sensitive screen that occurs subsequent to the time, wherein the previous touch event and the subsequent touch event occur within the time window, and wherein the classification and the re-classification contribute respective votes used to determine a final classification of the touch event.

2. The method of claim 1, wherein the respective votes contributed by the classification, the re-classification, and any other re-classification characterize the touch event as a palm touch or a pointer touch, and wherein the final classification is determined based on a type of classification of the touch event that has a majority of votes of a cumulation of the respective votes.

3. The method of claim 1, wherein the method further comprises:

presenting immediate feedback to a user if the touch event is classified as a pointer touch by rendering a stroke on the touch screen;

removing the stroke if a cumulation of the respective votes ever indicates that the touch event is to be classified as a palm touch.

4. The method of claim 1, wherein the features comprise statistics computed over a sequence of touch events, comprising the previous touch event, the touch event, and the subsequent touch event.

5. The method of claim 4, wherein respective touch events, including the touch event, the previous touch event, and the subsequent touch event, comprise a centroid and a radius, and wherein respective features of the respective touch events, including the series of features, the previous features, and the subsequent features, comprise a mean value, a standard deviation value, a minimum value, and a maximum value calculated for the radius of each touch event of the respective touch events.

6. The method of claim 5, wherein the respective features further comprise a total number of touch events in the sequence of touch events.

7. The method of claim 5, wherein the series of features further comprise Cartesian distances between the centroid of the touch event at the time the touch event was received and the centroids of other touch events of the respective touch events.

8. The method of claim 1, wherein the classifying is performed using a series of decision trees trained using the series of features or other features for varying window sizes of the time window, wherein the other features comprise the previous features and the subsequent features.

9. The method of claim 5, wherein the respective features further comprise a minimum distance to other touch events of the respective touch events and a number of touch events of the respective touch events.

10. The method of claim 1, wherein all touch events, including the touch event, the previous touch event, and the subsequent touch event, reported by the computing device are recorded and classified.

11. The method of claim 1, wherein the classification of the touch event occurs at approximately 25 ms after the touch event has been received.

12. The method of claim 1, wherein the re-classification occurs approximately 50 ms after the touch event has been received, and another re-classification occurs approximately 50 ms after the re-classification.

13. The method of claim 12, wherein the final classification occurs at a predetermined time or when the touch event terminates.

14. The method of claim 1, wherein the touch event lasts less than 25 ms, and wherein the final classification is based on data available at a termination of the touch event.

15. The method of claim 1, wherein the classifying comprises classifying the touch event upon receipt of the touch event.