HUMAN IDENTIFICATION USING AI

Surveillance systems frequently face challenges integrating advanced identification algorithms that extract speech or facial features from closed-circuit television (CCTV) camera feeds. However, gait analysis, which examines a person's distinctive walking pattern, stands out as a promising biometric modality due to its non-intrusive nature and effectiveness at a distance. This study proposes an automatic identification system that utilises gait analysis to recognise individuals from afar and predict the likelihood of matching between gait profiles. Leveraging computer vision, digital image processing, vectorisation, artificial intelligence (AI), and multi-threading, the system divides the gait cycle into four quarter-cycles to extract gait profiles from low-resolution camera feeds. Each gait profile is represented as four distinct vectors, totaling 80 features for each individual. The primary focus of this study lies in evaluating the speed-accuracy trade-off of the proposed model, which achieves a speed of 30 Hz and an average accuracy of 85% with a small training dataset. To assess its effectiveness compared to existing machine learning models, a receiver operating characteristic curve (ROC) is employed.