Читать книгу Human Motion Capture and Identification for Assistive Systems Design in Rehabilitation - Pubudu N. Pathirana - Страница 18

The Vicon [7] motion capture studio is used in some clinical settings as well as in certain other human motion assessment applications in sports. A number of fixed cameras within a dedicated infrastructure are used to capture the position of markers on the moving body part to greater precision, with the resulting Vicon system being used as the benchmark for motion capture systems. Due to cost, the requirement for a fixed and dedicated infrastructure and the specialised technical knowledge, this has not been used as a standard clinical practice and remains predominantly used in research and development.

Recently, a number of non‐invasive, portable and affordable optical 3D motion capture devices have emerged. These products include Leap Motion® controller, ASUS®Xtion PRO LIVE, Intel®Creative Senz3D, Microsoft Kinect®, and so on. Among them, Kinect is the most popular motion capture device for whole-body motion capture. The first version was released in 2010 with Xbox 360 and was used for gaming purposes and the second version was released with Xbox One in 2014.

The first version of Kinect utilised a depth sensor provided by a company named “PrimeSense” [400]. The appearance and components of this version of Kinect are shown in Figure 1.3.

Figure 1.3 Appearance and components of Kinect version 1. Source: Evan‐Amos, Image taken from https://commons.wikimedia.org/w/index.php?curid=16059165.

Figure 1.4 The pinhole camera model of Kinect version 1 [373]. Source: From Wang et al. [373]. © 2012, University of British Columbia.

The infrared projector and the corresponding camera can be modelled as in Figure 1.4.

These sensors measure the depth information via a structured light principle, which analyses a pattern (such as that in Figure 1.5) of bright spots projected on to the surface of an object [328]. In the case of Kinect, these “bright spots” are infrared light and are unobservable by human eyes directly.

Furthermore, Kinect utilised the other two techniques to further process the information to generate depth maps. These two tools include depth from focus and depth from stereo [121]. The principle of the former is that the further away the object is, the more blurred it will be [125], while the latter utilised parallax to estimate the depth information.

Different from the first version of Kinect, the second version (refer to Figure 1.6) measures the depth information with the time‐of‐flight (ToF) technique [189], which is stated as the distance that can be measured by knowing the speed of light and the duration the light uses to travel from the active emitter to the target. Estimated in Lachat et al. [189], this version of Kinect utilised the indirect time‐of‐flight, which measures the “phase shift between the emitted and received signal”. The depth is computed as

(1.1)

where f is the modulation frequency, c is the light speed and Δϕ is the determined phase shift.

Figure 1.5 An example of the projected pattern of bright spots on an object [328]. Source: Shpunt and Zalevsky [328].

Figure 1.6 Appearance of Kinect version 2. Source: Evan‐Amos, Image taken from https://commons.wikimedia.org/wiki/File:Xbox-One-Kinect.jpg.

As for the accuracy of joint positions tracked by the first version of Kinect, some studies have been done. Different studies have come to different conclusions on the accuracy of skeleton joint tracking. For instance, Webster et al. [376] reported that the accuracy was around 0.0275 m after removing offset by resetting the alignment of the average points for each record set. Therefore, they concluded that Kinect version 1 was sufficient for clinical and in‐home use. Obdrzalek et al. [263] evaluated the accuracy of the first version of Kinect, PhaseSpace Recap and Autodesk MotionBuilder in the environment of coaching elderly people. The result reported in their paper is that the error of the skeleton built by both Kinect and MotionBuilder is around 5 cm. However, in general postures, the accuracy is about 10 cm due to unavoidable factors, such as occlusions. Therefore, they suggested that the current skeletonisation approach enabled Kinect to measure general trends of movements, while an improved skeletonisation algorithm should be investigated if Kinect was used for quantitative estimations. Furthermore, Xu et al. [388] evaluated the accuracy of both the first and second versions of Kinect for static postures. From their experiment, they concluded that the accuracy of the first version Kinect varies from posture to posture. For instance, the error was only 26 mm for a shoulder centre in an upright standing posture, while it was 452 mm for the right foot joint in a sitting posture with the right leg on top of the left one. By comparison, for the second version of Kinect, when the right foot was raised, the error of the left elbow was only 26 mm, while it was 418 mm when the right leg was on the left one. As a result, it was concluded that though the resolution of the second version of Kinect had been improved significantly, its tracking accuracy of the joint centre had not improved.

The comparison of the specifications between the two versions of Kinects is shown in Table 1.3. From the comparison, it is obvious that the second version of Kinect provides a larger viewing angle, a higher resolution in both depth images and colour images, and more tracking joints.

Though Kinect was initially developed for gaming, it is widely applied in tele‐rehabilitation as a non‐invasive and affordable motion capture device. A telerehabilitation system (KiReS) using Kinect as the motion capture device has been proposed. On the patient side, two avatars were displayed to represent the motion recorded by the therapist (reference motion) and that performed by the patient. Therefore, the patient was able to see the differences between his/her motion and the reference. Eventually, the incorrect movements could be corrected over time. On the therapist side, new motions could be created to suit the patient's conditions by composing various existing movements or recording completely new ones. Luna‐Oliva et al. [217] utilised Kinect Sports I^TM, Joy Ride^TM and Disneyland Adventures^TM to provide telerehabilitation services to children with cerebral palsy in their school. Their experimental results showed that it is feasible to use Kinect as a therapeutic tool for children with cerebral palsy and the improvements in global motor function could be the result of using this tool. Ortiz‐Gutiérrez et al. [268] applied Kinect in providing telerehabilitation services to patients with postural control disorders. The experiment results showed an improvement over a general balance in both groups. In the experimental group, the significant differences resulted from visual preference and the contribution of vestibular information.