Accuracy evaluation of the Kinect v2 sensor during dynamic movements in a rehabilitation scenario

Abstract

In this paper, the accuracy evaluation of the Kinect v2 sensor is investigated in a rehabilitation scenario. The accuracy analysis is provided in terms of joint positions and angles during dynamic postures used in low-back pain rehabilitation. Although other studies have focused on the validation of the accuracy in terms of joint angles and positions, they present results only considering static postures whereas the rehabilitation exercise monitoring involves to consider dynamic movements with a wide range of motion and issues related to the joints tracking. In this work, joint positions and angles represent clinical features, chosen by medical staff, used to evaluate the subject's movements. The spatial and temporal accuracy is investigated with respect to the gold standard, represented by a stereophotogrammetric system, characterized by 6 infrared cameras. The results provide salient information for evaluating the reliability of Kinect v2 sensor for dynamic postures.

A review of computational approaches for evaluation of rehabilitation exercises

“…Comparable validation results were reported by Capecci et al [34], where for a set of three rehabilitation exercises an average positional RMSE of 3.3 cm (1.3 inch) and an average orientational RMSE of 12.7 degrees were recorded. In [88], Otte et al…”

“…In many related works, feature engineering is performed manually based on authors' understanding of human movements [33]- [37], [54], [97], [98], [126], [127]. For example, in [34], underarm angles and Euclidean distance between the elbows were used to describe the lifting of the arms. Similar, for squatting evaluation, knee angles and Euclidean distance between the ankles were extracted as clinical features.…”

Deep Learning-Based Upper Limb Functional Assessment Using a Single Kinect v2 Sensor

“…The limits of agreement (LoA) between the Kinect v2 sensor and the 3DMC system are 28 • , 46 • for peak knee flexion angle at a self-selected walking speed [15], 7 • , 25 • for trunk anterior-posterior flexion [16]. Average errors of 24 • , 26 • are observed for the right and left peak knee flexion angles during squatting [19].…”

“…The Kinect SDK v2.0 features skeletal tracking with 3D locations of 25 joints for each skeleton [12]. Kinect v2 has been employed in gait analysis [13][14][15], balance and postural assessment [16,17], foot position tracking [18], gait rehabilitation training [19,20], upper limb functional assessment or rehabilitation training [4,[21][22][23][24][25].Several studies have assessed the agreement between Kinect sensor and 3DMC. Kinect sensor demonstrated good reliability in assessing temporal-spatial parameters such as timing, velocity, or movement distance of functional tasks for both healthy subjects and people with physical disorders [4,13,21,22,26,27].…”

“…The Kinect SDK v2.0 features skeletal tracking with 3D locations of 25 joints for each skeleton [12]. Kinect v2 has been employed in gait analysis [13][14][15], balance and postural assessment [16,17], foot position tracking [18], gait rehabilitation training [19,20], upper limb functional assessment or rehabilitation training [4,[21][22][23][24][25].…”