What is it about?
Advances in Computer Vision, specifically Human Pose Estimation (HPE) methods have enabeled the application of such methods even to visually challenging scenarios such as a jiu-jitsu grappling match. For Brazilian jiu-jitsu (Bjj) as well as other sports and martial arts HPE approaches are especially interesting as the poses of the athletes contain most of the information about a match. If we are able to detect the poses of the athletes from an image and reliably track them, we can obtain information of athletes movement simply by processing a video of the sport activity. This data enables many practical applications. In our paper we use the detected and tracked poses of the athletes to classify the positions of the athletes and based on the classification results automatically score a match.
Featured Image
Photo by Richard Bustos on Unsplash
Read the Original
This page is a summary of: Video-Based Detection of Combat Positions and Automatic Scoring in Jiu-jitsu, October 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3552437.3555707.
You can read the full text:
Resources
Brazilian Jiu-jitsu Dataset
The Brazilian Jiu-Jitsu Positions Dataset is a dataset for training jiu-jitsu position classification methods. The dataset contains 120.280 labeled images of 2 jiu-jitsu athletes sparring in different combat positions.
Automatic Jiu-jitsu Scoring and Combat Position Prediction
A jiu-jitsu sparring sequence with pose detections, athlete tracking, position predictions, and automatic scoring. Supplementary material of the paper Video-Based Detection of Combat Positions and Automatic Scoring in Jiu-jitsu. Position predictions can be seen in the bottom left corner, with correct predictions in green text, wrong predictions in red text and hard to label positions in blue text. Automatically detected and actual results can be seen in the top right corner.
Contributors
The following have contributed to this page
