In this article, we present a large 3D motion capture dataset of Taijiquan martial art gestures (n = 2200 samples) that includes 13 classes (relative to Taijiquan techniques) executed by 12 participants of various skill levels. Participants levels were ranked by three experts on a scale of [0-10]. The dataset was captured using two motion capture systems simultaneously: 1) Qualisys, a sophisticated optical motion capture system of 11 cameras that tracks 68 retroreflective markers at 179 Hz, and 2) Microsoft Kinect V2, a low-cost markerless time-of-flight depth sensor that tracks 25 locations of a person׳s skeleton at 30 Hz. Data from both systems were synchronized manually. Qualisys data were manually corrected, and then processed to complete any missing data. Data were also manually annotated for segmentation. Both segmented and unsegmented data are provided in this dataset. This article details the recording protocol as well as the processing and annotation procedures. The data were initially recorded for gesture recognition and skill evaluation, but they are also suited for research on synthesis, segmentation, multi-sensor data comparison and fusion, sports science or more general research on human science or motion capture. A preliminary analysis has been conducted by Tits et al. (2017) [1] on a part of the dataset to extract morphology-independent motion features for skill evaluation. Results of this analysis are presented in their communication: "Morphology Independent Feature Engineering in Motion Capture Database for Gesture Evaluation" (10.1145/3077981.3078037) [1]. Data are available for research purpose (license CC BY-NC-SA 4.0), at https://github.com/numediart/UMONS-TAICHI.

In this paper a methodology to recognize actions based on RGB videos is proposed which takes advantages of the recent breakthrough made in deep learning. Following the development of Convolutional Neural Networks (CNNs), research was conducted on the transformation of skeletal motion data into 2D images. In this work, a solution is proposed requiring only the use of RGB videos instead of RGB-D videos. This work is based on multiple works studying the conversion of RGB-D data into 2D images. From a video stream (RGB images), a two-dimension skeleton of 18 joints for each detected body is extracted with a DNN-based human pose estimator called OpenPose. The skeleton data are encoded into Red, Green and Blue channels of images. Different ways of encoding motion data into images were studied. We successfully use state-of-the-art deep neural networks designed for image classification to recognize actions. Based on a study of the related works, we chose to use image classification models: SqueezeNet, AlexNet, DenseNet, ResNet, Inception, VGG and retrained them to perform action recognition. For all the test the NTU RGB+D database is used. The highest accuracy is obtained with ResNet: 83.317% cross-subject and 88.780% cross-view which outperforms most of state-of-the-art results.