Research On Handwriting Recognition Based On Smart Watch

Nowadays smart wearable devices(such as smart bracelets,smart wristbands,and smart watches)have gradually blending into individuals' daily lives,and have become an indispensable intelligent hardware for people.They can not only record daily activity information,such as steps,sleep quality,calorie consumption,heart rate,etc.,but also provide people with a new means of human-computer interaction.And the gesture recognition functions have been added to these devices,such as raising the wrist to brighten the screen and flipping the wrist to switch the interface.The proposed handwriting recognition technology based on the smart watch of this paper focuses on real-world scenarios where users wear smart watches for information recording.It can effectively complement existing fine-grained action recognition methods.This paper proposes a handwriting recognition scheme based on a accelerometer and a gyroscope with a in-depth research of text input methods for wearable devices.This scheme is dedicated to the research of finegrained handwriting recognition under a cross-user condition.The main contribution is listed as follows:(1)Made full use of the advantages of deep learning models in extracting high-level semantic features.A multimodal CNN(Convolutional Neural Network)network is designed for feature extraction of the sequential data,and the implicit expression inside the single-modality is abstracted in a layered manner.And multimodal fusion is performed to build a feature extraction model with stronger ability to mine deep information.(2)Proposed the bidirectional LSTM(Long Short-Term Memory)network to model time series,which takes advantages of temporal dependencies and semantic information to better predict output.And this paper introduces the CTC(Connectionist Temporal Classication)algorithm into the sensor data-based action recognition,effectively avoiding the pre-segmentation in traditional methods which leads to a drop in accuracy and enabling an endto-end identification mechanism.(3)Designed a series of experiments to verify the recognition performance of the smartwatch-based handwriting recognition.At the same time,the main factors affecting the system performance are analyzed in detail,and then the prototype system is implemented using a commercial Ticwatch.This paper first describes the background and significance of handwriting recognition research based on smart watches,analyzes the shortcomings in existing research,then gives a brief overview of sensor-based motion recognition technology,and details the technical details of each module of the proposed handwriting recognition scheme.After that,a prototype handwriting recognition system is designed and implemented.Finally,a detailed experimental evaluation and a experimental analysis are carried out,the research work is summarized and the future work is prospected.