Technology

Motion

AI technologies, long researched to emulate and learn from humans in text, speech, imagery, video, and 3D data, have now advanced to recognizing and analyzing human behavior (motion). In addition to image and video data, we continuously acquire motion training data through our in-house motion capture studio to drive research and develop solutions in the motion domain

2D Pose estimation

  • Technology for inferring 2D joint coordinates of the human body from images
  • Applied in sports analytics, action recognition, AR/VR, and more
  • Proven capability to accurately estimate poses for specialized movements, such as Taekwondo techniques
  • Dataset generation for specialized motions using our motion capture studio
3D 자세 추정

3D Pose estimation

  • Estimate 3D human poses using single or multiple camera setups
  • Leverage our motion capture studio to generate proprietary 3D datasets in addition to open-source data for model development
  • Deliver a significant advantage in 3D model training by overcoming the severe limitations of manual 3D data labeling
동작인식

Action recognition

  • Motion recognition is broadly categorized into skeleton-based data or RGB-based image data for action recognition
  • Action recognition model selection is determined by the task, and we possess expertise in both approaches
  • By combining object detection and action recognition, we can identify each object’s behavior type within a single frame
  • We hold one registered patent for Taekwondo action recognition