Evaluating Pointing Accuracy on Kinect V2 Sensor

In this short paper we present an accuracy analysis of the new Kinect V2 depth sensor solely used as a pointing device based on 3D joint positions of the user’s arm. Computation of the pointing vector has been done in two ways. First calculated from elbow and wrist position and then from shoulder and wrist position. Based on the evaluation of Fitts’ Law we have examined the sensor capabilities with a group of 10 computer affine participants. Results show, that shoulder-wrist based pointing vector calculation performs better (24.3% in average) than elbow-wrist based calculation for clickable targets varying from 39cm to 4.6cm in diameter ordered in distances varying from 28cm to 79.5cm and clicked from a distance of 3.1m. Furthermore we propose that the sensor should be positioned near the ground plane to reduce detection instabilities. Since Kinect V2 preview hardware and API is preliminary and subject to change, we will re-evaluate our findings after official hardware release.