Abstract:
In the field of computer vision, person identification across several cameras is a very important task especially the identification of the same person in different scenes. Human gait identification is a biometrics technique that permits identifying an individual from a long distance. Gait identification is based on the human walking style in gait recognition many covariate factors can affect the performance of gait which includes different dresses, carrying bags, long coats, coating shoes, bag packs, and different angles of view. Clothing and carrying bags are the covariates that affect the accuracy of the gait identification system. We proposed a method that solves the challenge of different covariate factors. In gait recognition, the first pose estimation network are used pose estimation network gives the probabilities of different body joints and the pose estimation network gives pose coordinate sequences and these coordinates are preprocessed we normalize and selects the effective joint in order to extract more robust features After this pose coordinates are passed to the gait network. Convolutional neural networks are used to convert spatial features into 1-dimensional pose descriptors. After this one-dimensional vector is fed into the LSTM network.LSTM is used to extract temporal features. All temporal features are finally aggregated with Average temporal pooling into a one-dimensional identification vector with good discriminatory properties. SVM classifiers are used for classification purposes we used the tum gait dataset for training and testing
