Model-based person identification in multi-gait scenario using hybrid classifier

In recent decades, gait has become an important topic in biometric. Gait attains popularity in human authentication because of its non-cooperation and sensing of gait patterns from a distance. However, in the real-time environment, recognizing based on gait is a challenging task when multiple people walk in a group. Therefore, this paper focused on recognizing people in a multi-gait (MG) scenario. Multi-gait means that more than one person are walking in a group. Here, our work is divided into two phases. In the first phase, we reconstruct the occluded regions. Here, we present a model-based approach to recognize a person in a multi-gait scenario. Therefore, five dynamic regions of interest (ROIs), such as Ankle, Knee, Wrist, Elbow, and Shoulder, are taken, and a numerical interpolation approach is applied to regenerate the occluded ROIs. Then, in the second phase, we extract linear kinematic features and propose a hybrid classifier for model-based multi-gait identification. Finally, the experimental results of the hybrid classifier, i.e., PSO-NN, demonstrate that the proposed classifier performs better in multi-gait identification than the state-of-the-art classifier, such as k-NN and ANN.