MACHINE-LEARNING BASED TAP DETECTION

摘要

An electronic device can be configured to enable a user to provide input via a tap of the device without the use of touch sensors (e.g., resistive, capacitive, ultrasonic or other acoustic, infrared or other optical, or piezoelectric touch technologies) and/or mechanical switches. Such a device can include other sensors, including inertial sensors (e.g., accelerometers, gyroscopes, or a combination thereof), microphones, proximity sensors, ambient light sensors, and/or cameras, among others, that can be used to capture respective sensor data. Feature values with respect to the respective sensor data can be extracted, and the feature values can be analyzed using machine learning to determine when the user has tapped on the electronic device. Detection of a single tap or multiple taps performed on the electronic device can be utilized to control the device.

主权项

1. A computer-implemented method for detecting a tap as input for an electronic device, comprising:
capturing a sensor data vector corresponding to a one-dimensional signal from at least one of an accelerometer or a gyroscope; calculating an integral buffer corresponding to the sensor data vector by summing: (a) a sensor signal value at each position in the sensor data vector corresponding to the position of the integral buffer, and (b) sensor signal values, prior to the position, in the sensor data vector; calculating a feature value vector using: (a) the integral buffer, and (b) a plurality of one-dimensional Haar-like features comprising sequences of numbers of equal length including a first sequence of one or more negative ones and a second sequence of one or more positive ones; for a decision tree classifier of a random forest classifier that is trained to detect that the tap has been performed on a back surface of the electronic device, evaluating a respective subset of feature values of the feature value vector with respect to the decision tree classifier to obtain a respective classification; and determining that the tap has been performed on the back surface of the electronic device based on the respective classification obtained from each decision tree classifier of the random forest classifier.