Trajectory recovery method of handwritten characters in the air based on deep adversarial learning in feature space

Abstract

The invention discloses a method for recovering the trajectory of handwritten characters in the air based on feature space deep confrontation learning. The steps are as follows: perform noise reduction filtering on the acquired inertial sensing signal sequence and plane trajectory coordinate sequence, and divide the training sample set and the test sample set. ; Design a domain migration model based on feature space deep confrontation learning; train the model with a training sample set; input the inertial sensing sequence in the test data into the trained domain migration model, and the sequence output by the model is used as the result of trajectory recovery. The present invention realizes cross-domain sample migration based on deep confrontational learning in feature space, and can convert inertial sensing signals of handwritten characters in the air into a plane trajectory coordinate sequence to realize visualization of inertial sensing signals; no paired inertial sensing signals are required Sequence and planar trajectory coordinate sequence can also complete model training, and the symmetrical structure can realize two-way migration between inertial sensing signals and planar trajectory, and the trajectory recovery accuracy and smoothness are good.

Description

technical field [0001] The invention relates to the technical fields of deep learning and artificial intelligence, in particular to a method for recovering handwritten character trajectories in the air based on feature space deep confrontation learning. Background technique [0002] Air handwriting based on inertial sensors (accelerometers and gyroscopes) is one of the emerging cutting-edge research directions in the computer field in recent years, and has a wide range of applications in smart home, automatic driving, education, medical care, industrial production, and assisted communication. However, inertial signals (acceleration and angular velocity signals) are quantities that measure changes, and their readability is poor. It is difficult to identify specific written content from the waveform only by human observation. This feature of the inertial signal brings difficulties to bad sample cleaning, sample labeling, and character-level sample segmentation in the data proc...

AI Technical Summary

Problems solved by technology

The traditional method focuses more on recovering the trajectory directly from the inertial signal and correcting the cumulative error of the motion trajectory, lacking the effective use of the real writing trajectory, and the recovery accuracy and practicability are increasingly unsatisfactory to users

On the other hand, supervised domain transfer methods require paired inertial sensing signals and planar trajectory samples. At present, datasets of handwritten characters in the air containing paired samples are very scarce, which brings difficulties to model training.

Images