Gait recognition method based on GRU

Abstract

The invention discloses a gait recognition method based on a GRU, and belongs to the technical field of prostheses. According to the invention, the problems of complex calculation, poor real-time performance and the like of a traditional gait classification method are solved; complex feature extraction engineering is omitted through the GRU, only model parameters need to be used for classification, the calculation speed is greatly increased, real-time calculation of the gait stage is achieved, and the tedious process that traditional gait recognition needs to be classified offline is eliminated. The method comprises the following steps: acquiring plantar pressure information during walking by using an FSR film pressure sensor worn on the plantar of an artificial limb; marking correspondinglabels on the data queues according to the target typical walking characteristics and the timestamps; building a GRU network model; defining a GRU unit, a full connection layer and each activation function; and dividing the obtained data label pair into a training set and a test set, sending the training set to a GRU network model for training, evaluating a model classification effect by using the test set after training is completed, and performing online real-time classification.

Description

technical field [0001] The invention belongs to the technical field of artificial limbs, and in particular relates to a GRU-based gait recognition method, that is, a lower limb prosthetic gait recognition system, which can recognize the gait stages of a prosthetic wearer and improve the accuracy and accuracy of gait recognition. real-time. Background technique [0002] A complete cycle of gait is called a "gait cycle". A gait cycle is divided into two phases, the "stance phase" and the "swing phase", which can be further divided into sub-phases. Gait recognition can not only provide an important analysis basis for rehabilitation physicians, but also provide control signals for intelligent prosthetics to make corresponding control strategies and parameter adjustments, so that the patient's movement process is more stable, smooth and natural. [0003] Most of the current gait recognition methods have disadvantages such as complex calculation, poor real-time performance, and ...

AI Technical Summary

Problems solved by technology

[0003] Most of the current gait recognition methods have disadvantages such as complex calculation, poor real-time performance, and low reliability.

For example, the Chinese invention patent application No. 201811241695.7 is to identify the gait stage by collecting EMG signals and extracting relevant features. However, the detection signal is affected by factors such as surface temperature and sweat, and the stability and accuracy are low. Wavelet decomposition and calculation are required. The complex feature extraction process such as Willison amplitude is cumbersome and has poor real-time performance, and the electrodes need to be in direct contact with the skin, which is not comfortable

The Chinese invention patent application No. 201910976122.7 uses the IMU module to collect the rotation angles of the left and right thighs and calves of the human body, and uses a rule-based classification algorithm to realize real-time recognition of the human walking gait. However, the transition period of the gait stage is short and the signal changes are complicated. , there are large differences in different road conditions and different detection objects. Although the rule-based classification algorithm is simple to calculate, its classification accuracy is doubtful and its reliability is low.

Images