Ownership Estimation for Tracked Hands in a Colocated VR Environment

Abstract

Hand tracking systems play a crucial role in virtual reality (VR) applications, typically focusing on tracking the hands of the user who is using the system. Consequently, most existing systems are designed to track a maximum of two hands simultaneously. However, in certain colocated multi-user VR scenarios, it becomes necessary to track more than two hands simultaneously, such as to eliminate blind spots in individual tracking systems. In such scenarios, accurately assigning the tracked hands to the corresponding users using only the hand locations relative to the users becomes essential. This paper introduces and evaluates various methods for efficiently assigning hands to users in such scenarios. Additionally, we propose an algorithm that leverages past assignments to enhance the robustness and effectiveness of future assignments. Our experimental results demonstrate that this algorithm significantly improves upon existing methods. Furthermore, when combined with an assignment algorithm based on reinforcement learning AI agents, we achieve a remarkable 99% accuracy in hand assignments. As a result, we present an assignment algorithm specifically tailored for colocated VR scenarios, utilizing only the hand and user locations within the scene, making it directly applicable in the aforementioned contexts.