An Application of Evaluation of Human Sketches using Deep Learning Technique

Abstract

This research is a study of the evaluation of full-body sketches and the principle of the human pose estimation using the OpenPose library, a method to detect 18 keypoints on a human structure. The dataset used in this research was drawing sketches of 22 first-year students, each of whom drew three drawings of three models. Detected keypoints are calculated to determine the angle and distance between keypoints, which provides 26 features. These features were modeled using ANN for predicting the grades of drawings classified as good, moderate, poor. The resulting keypoints are then taken to find the angles and distances of the skeleton, extracting 26 features and taking these features to create a model using ANN classification. The performance of the model was evaluated using with 56% accuracy © 2021 ACM.