Evaluation of 2D and 3D posture for human activity recognition (original) (raw)

Recognition of human activities is one of the most significant fields in the area of machine intelligence research, with its main objective of distinguishing various human activities. Accurate human posture extraction from images is a crucial step towards this objective. In this paper, laboratory dataset consisting of 2D and 3D images of subjects performing seventeen different activities have been collected, and then used to construct 2D and 3D human postures. Eight different classification models are subsequently used to classify the different activities. Subsequently, it has been shown that Random Forest classification model gives the best performance, in terms of accuracy. The findings also demonstrate that both 2D and 3D postures are capable of achieving significant accuracy score, with very high average accuracies of 99.18% and 99.82%, respectively, using the Random Forest classification model.