Metrological evaluation of KinectFusion and its comparison with Microsoft Kinect sensor

Abstract In recent years the use of low cost sensors like Microsoft Kinect in the engineering field is growing day by day. Its use as a measurement device, capable of generating geometric data in close-range environment has been already validated. KinectFusion is a software that is able to capture a complete dataset of a 3D scene reconstruction using the Microsoft Kinect depth sensor at a very low cost. In this work, a metrological geometric verification of the KinectFusion implementation is performed using a standard artefact which consists of five delrin spheres and seven aluminum cubes. Accuracy and precision tests at different resolutions and ranges were performed to assess the metrological behavior of KinectFusion and compare it with the Microsoft Kinect depth sensor. The analysis reveals that both approaches yield a decrease in precision and accuracy with increasing range between sensor and object. Measurement of the length between the centers of the spheres was used for accuracy assessment. Taking into account the farthest sphere, an increasing trend of the accuracy with the distance was obtained for KinectFusion dataset, independently of the 3D model resolution and the range. By contrast, this trend is not shown for Kinect depth sensor dataset. KinectFusion precision results are better than the obtained with Kinect stand alone sensor. Besides that, precision values for KinectFusion are proportional to the resolution of the 3D model. KinectFusion algorithm provides a suitable 3D model of the scene that can be used for engineering applications where accuracy requirements are not very strict.