MODELS AND METHODS FOR GAIT-BASED PERSON IDENTIFICATION WHILE WEARING DIFFERENT OUTFIT

Abstract

The purpose of this study is to present a technique for detecting a person based on their gait. This approach is designed to be resistant to changes in clothing and to provide high identification accuracy rates even when subjects wear various kinds of apparel. Specifically, the technique is based on the exploitation of the suggested LR2P descriptor, which demonstrates resilience to rotation and allows the extraction of more discriminative characteristics, particularly when participants wear a variety of apparel. The OUISIR B and CASIA B datasets were used in an experimental investigation that was carried out in order to verify the efficiency of the approach that was specifically devised. For every dataset, there were two separate experiments that were made. It was discovered that the suggested technique achieved an accuracy rate of 86.12% when applied to the OUISIR B dataset, which resulted in beneficial outcomes for a variety of clothing sets. Furthermore, the second experiment conducted on this dataset indicated that the suggested strategy efficiently mitigates the problem of differences in clothing and maintains a high identification rate, even when such changes are large. This was proved throughout the results of the trial. There were 32 different tests, and the average recognition accuracy that was attained was 92.02% each. The experiment that was carried out on the CASIA B dataset demonstrated that, when compared to the outcomes of various other recognition techniques that were evaluated on the CASIA B gait dataset, the suggested approach outperformed them in the majority of situations. In addition, the second experiment, which was conducted on the CASIA B dataset, demonstrated the promising outcomes of the suggested feature extraction approach.