Evaluating the Immediate Applicability of Pose Estimation for Sign Language Recognition

  • 2021-04-20 14:41:45
  • Amit Moryossef, Ioannis Tsochantaridis, Joe Dinn, Necati Cihan Camgöz, Richard Bowden, Tao Jiang, Annette Rios, Mathias Müller, Sarah Ebling
  • 1

Abstract

Signed languages are visual languages produced by the movement of the hands,face, and body. In this paper, we evaluate representations based on skeletonposes, as these are explainable, person-independent, privacy-preserving,low-dimensional representations. Basically, skeletal representations generalizeover an individual's appearance and background, allowing us to focus on therecognition of motion. But how much information is lost by the skeletalrepresentation? We perform two independent studies using two state-of-the-artpose estimation systems. We analyze the applicability of the pose estimationsystems to sign language recognition by evaluating the failure cases of therecognition models. Importantly, this allows us to characterize the currentlimitations of skeletal pose estimation approaches in sign languagerecognition.

 

Quick Read (beta)

loading the full paper ...