Selection Intent Prediction in Online Virtual Environments : A Comparison Study

Sammanfattning: In this work, we address the problem of the selection of 3D objects based on the user’s intent to interact with them in online virtual environments, both on desktop and VR devices. In addition to the traditional ray-casting technique, we select four promising techniques from the literature using time-dependent volumetric scoring (IntenSelect), trajectory prediction (KEP), Kalman filters enhancement (KalmanKEP) and machine learning with decision trees (MTP). We refine and adapt them to be implemented on both devices. We gather a number of raycasting-based 3D selections in a remote virtual environment, forming a small dataset that serves for simulations and training. The performance of the techniques is evaluated using these simulations through the theoretic relative amount of time and movement required for each technique to be able to predict the target corresponding to the user’s selection intent, called availability. The users’ experience is assessed through a user evaluation on two axes that are usability and task load. The results show that IntenSelect and the adapted KEP technique are credible alternatives to ray-casting. However, the ray-casting technique is still among the preferred ones in terms of user experience, despite being a method with a low spatial and temporal availability compared to others.