Detecting the Point of Release of Virtual Projectiles in AR/VR

Abstract

Our aim is to detect the point of release of a thrown virtual projectile in VR/AR. We capture the full-body motion of 18 participants throwing virtual projectiles and extract motion features, such as position, velocity, rotation and rotational velocity for arm joints. Frame-level binary classifiers that estimate the point of release are trained and evaluated using a metric that prioritizes detection timing to obtain a ranking of joints and motion features. We find that wrist joint and rotation motion feature are most accurate, which can can help when placing simple motion tracking sensors for real-time throw detection.