| Title |
Body KeyPoint Detection and Pose Similarity Measurement for Home Training |
| Authors |
강도희(Do-hee Kang) ; 이준민(Jun-min Lee) ; 송병철(Byung Cheol Song) |
| DOI |
https://doi.org/10.5573/ieie.2021.58.8.98 |
| Keywords |
Key points detect; Feature extractor; Pose estimation; Auto encoder; Pose similarity |
| Abstract |
Home training is drawing attention under the development of a ‘Untact Culture’ along with an interest in personal health. As a result, many home training applications provide posture similarity measurement based on body core point detection. However, it has limitations in that its own contents are available. To compensate for this point, in this paper, we propose an application to measure real-time pose similarity on target images freely set by the user. The proposed application combines AlphaPose, a deep learning-based body-core point detector, and learned autoencoders to extract latent vectors featuring postural features. Then, the similarity between the two postures is scored through a radial basis function (RBF) and provided to the user. The method of pose similarity measurement used in this application selected the most appropriate similarity comparison metric by comparing and analyzing the commonly used metrics in existing Action Recognition tasks or Anticipation tasks, such as AP(average precision), cosine similarity. The proposed application maximizes the effectiveness of home training and increases usability. |