Motion behavior identification method based on LSTM model

Abstract

The main technical problem to be solved in the invention is to provide a motion behavior identification method based on an LSTM model, which can achieve high identification accuracy with relatively less data, overcomes the shortcoming that the current classification algorithm needs manual feature extraction, and can be widely applied in practice. The method comprises the following steps: collecting time series data related to multiple groups of motion of different people through sensor bracelets worn on the hands of the people; S2, marking motion types corresponding to the multidimensional raw data collected to ensure subsequent supervised learning; S3, carrying out necessary treatment on training data, and transmitting the training data as input data to an LSTM model for training to get the best neural network parameters, and taking the model as a final identification model; and S4, preprocessing to-be-identified motion behavior data, calculating a motion sequence with maximum probability in the output layer by taking the to-be-identified motion behavior data as the input of the LSTM model, and taking the result as the motion type finally identified.

Description

technical field [0001] The invention relates to the field of human motion recognition in pervasive computing, more specifically, a motion behavior recognition method based on an LSTM model. Background technique [0002] Ubiquitous computing is also known as ubiquitous computing and pervasive computing. This concept emphasizes computing integrated with the environment, while the computer itself disappears from people's sight. Under the mode of ubiquitous computing, people can acquire and process information anytime, anywhere and in any way. Discontinuous connection and lightweight computing (that is, relatively limited computing resources) are the two most important features of ubiquitous computing. The software technology of ubiquitous computing is to realize transaction and data processing in this environment. [0003] Early motion recognition was mainly based on vision, given an image sequence or a video clip, to identify the type of motion of a person. The vision-based...

AI Technical Summary

Problems solved by technology

[0005] The technical problem mainly solved by the present invention is to provide a motion behavior recognition method based on the LSTM model, which can obtain a good recognition accuracy rate with relatively less data, and overcome the shortcomings of manual extraction of features in the current classification algorithm, so that it can widely used in practice

Images