A NEW DEEP LEARNING ALGORITHM FOR VIDEO BASED GAIT BASED RECOGNITION USING HYBRID DEEP CONVOLUTIONAL NEURAL NETWORKS
This paper describes a way to solve the problem of biological recognition based on the patterns of motion made by a person walking. The suggested method uses the data collected by the accelerometer and gyroscope sensors of a smartphone while the user is doing the gait action to improve the design of a recurrent neural network (RNN) to learn the features that best describe each person. The database has 15 people, and the acceleration data is given in a format with three directions (X, Y, and Z). Data are handled ahead of time to determine the motion in the direction of gravity. For person identification, a deep recurrent neural network model is used. This model is made up of LSTM cells that are split into several layers and thick output layers. Most of the time, the end design gives answers that are more accurate than 97%. The suggested design based on deep neural networks is tried in different situations to see how well it works and how well it holds up.
Accelerometry; Gait; Walk; Identification; Recognition; Recurrent Neural Network; LSTM; Accuracy; Smartphone