영상기반 관절 각도 특징 분석을 이용한 사람 식별

Abstract

In this study, the angles between gait-relevant joints are considered a robust and differential feature set. The aim of this paper is to develop an approach that identifies a person’s gait cycle using body-joint information. The proposed approach acquires six different joint angle measurements using an RGB depth sensor, and then stores these in a queue-attribute collection. A genetic algorithm is then applied to reduce the number of features from 120 to 43. Following this, the data is trained with both a random forest classifier (RFC) and a K-nearest neighbor (KNN) algorithm . An average accuracy of 95.64% and 91.98% for gait analysis and identification with RFC and KNN algorithms respectively.