Research On Motion Trajectory Recognition Of Smart Phone Based On LSTM

The motion track recognition technology based on inertial sensors is a new type of gesture recognition technology that has emerged in recent years.With the advantages of strong anti-interference ability and scalability,it has a wide application prospect in many fields,especially in the field of human-computer interaction.The deep learning model can construct the recognition model while automatically extracting the key features of the data.It has become an effective method to improve the performance of the recognition system.In recent years,it is a research hotspot in the field of motion track recognition based on inertial sensors.A large number of theoretical studies on trajectory recognition technology of inertial sensor motion are the first to manually extract key features from inertial sensor signals,and then establishes a recognition model based on these characteristics to realize motion trajectory recognition.However,influenced by factors such as sensor materials,manufacturing processes,and operational errors,the acquired inertial sensor signal components are complex.Artificial feature extraction method is often difficult to extract good key feature information,which leads to the low recognition rate of traditional recognition models.In order to solve the problem of difficulty in the extraction of artificial features and the low recognition rate of traditional models,the focus is on the use of deep learning technology to realize the recognition of the trajectory of smart phones.Aiming at the difficulty of extracting artificial features and the low recognition rate of traditional models,this paper focuses on the realization of the motion trajectory identification of smart phones using deep learning technology.The main research work includes:(1)Based on the acquisition of smart phone motion trajectory data,a series of preprocessing operations are performed on these data: data elimination,format conversion,data normalization,normalization and removal of gravity acceleration components.A motion trajectory data set based on the acceleration sensor of a smartphone and a trajectory data set based on a smartphone acceleration sensor and a gyroscope are constructed,and up to 5,500 sample data per dataset.(2)The Tensor Flow framework was used to establish the LSTM(Long Short-Term Memory)model.Trained the model and adjused model parameters using the two constructed data sets,the L3-50 model based on the trajectory recognition of the smart phone acceleration sensor and the L5-70 model based on the smart phone acceleration sensor and the gyro motion trajectory recognition were obtained.(3)Using the smart car as the experimental platform,the human-vehicle interaction system was constructed using the smart phone motion trajectory recognition model,and a dual navigation system capable of real-time switching of the navigation mode was realized.The visually-based autonomous navigation mode is capable of performing the navigation task in the general environment.,it can effectively overcome the environmental problems such as backlight shadow,light imbalance and other issues,manual navigation mode based on the trajectory of the smart phone can be competent for navigation tasks in a complex environment.The combination of the two navigation modes works well and has a certain application value for the research of smart vehicle assisted driving systems.