Interchangeability of Kinect and Orbbec Sensors for Gesture Recognition

In the context of recent advances and high market dynamics of the 3D motion camera products widely used in physical rehabilitation systems, we evaluate in this study the similarity and interchangeability of a first generation motion sensor, Microsoft Kinect, with one of the latest generations sensors for skeletal tracking, Orbbec. We have analyzed several body postures represented by skeletal points with each sensor, and the possibility to mix or interchange data from the two sensors to train machine learning algorithms and accurately classify these body postures. We have analyzed the performance of 23 of the best performing classifiers with skeletal data, using for testing our own data collected with Kinect and Orbbec. Results show there is a similarity between the two sensors and the possibility to interchange the sensor data, with no impact on accuracy for the majority of the classifiers used (16 out of 23), having minor adaptations applied on the data. The accuracy for the two sensors is similar when used independently. This proves it is possible to drive continuity between these two different generations of 3D motion cameras in systems performing gesture recognition.