Abstract—Visual-based human action analysis is an important research topic in the field of computer vision, and has great application prospect in intelligent sports. Home-based fitness is increasingly common in recent years, however lacking of accurate feedback and scientific guidance main easily lead to problems such as exercise injuries. In this paper, we propose an analysis system for Tai Chi action quality assessment and visual analysis with a consumer RGB-D camera. The main innovative work is as follows: 1. for home-based fitness action evaluation, we design a real-time intelligent analysis system combined with expert rules through a consumer RGB-D camera; 2. we transform the evaluation of 24-form Tai Chi Chuan into an artificial intelligence (AI) model, and realize action recognition and assessment through computer vision; 3. to train the AI model, we build a new dataset named TaiChi-24, which contains 1,408 samples with RGB-D images and 3D skeletons. We carry out evaluation experiments and analyses, and the experimental results have shown the advantage of applying our evaluation method on the proposed TaiChi-24 dataset.

If someone wants to download the TaiChi-24 dataset, please fill in the agreement, and email Kehao Ran<rkh117@bsu.edu.cn> or Jianwei Li <jianwei@bsu.edu.cn> to request the download link.