Virtual reality in the serve of motion analysis: a solution for fusing skeletal representation data from multiple Kinect devices

Background:Analysis of human skeleton data in motion can be used to provide feedback on the incorrect body posture or incorrect sequence of physical exercises. This information is mainly used to improve the ergonomic standards of the environment especially for people with restricted mobility. The emergence and development of Microsoft Kinect technology in the last decade, has opened a new way to detect human motions in real time. However, the low accuracy of single Kinect devices hinders its further development in rehabilitation and sports medicine. In order to achieve higher accuracy and usability for smart health applications, we propose a practical framework for human skeleton tracking and analysis that performs the fusing of skeletal data from multiple Kinect devices to provide a complete coverage of a subject.Material and Methods:The calibration process and data fusion algorithm are described with algebraic operations in vector space for deploying three Kinect units. The performance of human skeleton during motion activities was evaluated by several types of quantitative metrics including amplitude, velocity and position of marked joints. The experimental data was provided by 10 healthy subjects (7 males and 3 female) with no reported motor disorders. The accuracy of recognition was calculated for several standard and non-standard postures. To analyse the dynamic characteristics of skeleton motion during exercise, the speed of joints was computed as the distance travelled in the time interval. The asymmetries in the joint movement amplitudes and speed between the left and right side of the body were monitored to determine the correctness of execution of training sequence. The reliability of the obtained results has been validated using descriptive statics and test-retest method followed by Tukey mean-difference statistical analysis.Results:The proposed algorithm could successfully fuse the information of three Kinects regardless of their orientation angle and produce a much better tracked representation of users than single Kinect. Moreover, the proposed setup could monitor the evolution of joints during motor tasks. The results showed comparatively low average error rates around 10±5.4% for posture recognition. Our results are in-line with the results achieved by other authors: The intra-session test–retest reliability of the results expressed by ICC> 0.76 (excellent), R-squared> 0.8 (substantial) and CoV> 0.011(acceptable). Conclusion:This study contributed towards the solution of multi-sensor data fusion problem. Moreover, calculating quantitative measures indicators such as the position of joints, speed of movement, body asymmetry and rate of fatigue provided more accurate view on physical human performance during exercise. Consequently, it can be inferred that multi-Kinect based monitoring systems have the potential to assist patients in adjusting more suitable training program or even conducting home-based rehabilitation

کلیدواژه ها:

Multiple Kinects ، Rehabilitation ، Data fusion ، Sports medicine ، Human skeleton tracking