Kinematics Analysis Multimedia System for Rehabilitation

Abstract

Abstract. Driven by recent advances in information and communications technology, tele-rehabilitation services based on multimedia processing are emerging. Gait analysis is common for many rehabilitation programs, being, for example, periodically performed in the post-stroke recovery assessment. Since current optical diagnostic and patient assessment tools tend to be expensive and not portable, this paper proposes a novel marker-based tracking system using a single depth camera which provides a cost-effective solution that enables tele-rehabilitation services from home and local clinics. The proposed system can simultaneously generate motion patterns even within a complex background using the proposed geometric model-based algorithm and autonomously provide gait analysis results using a customised user-friendly application that facilitates seamless navigation through the captured scene and multiview video data processing, designed using feedback from practitioners to maximise user experience. The locally processed rehabilitation data can be accessed by cross-platform mobile devices using cloud-based services enabling emerging tele-rehabilitation practices.

A Depth Camera Motion Analysis Framework for Tele-rehabilitation: Motion Capture and Person-Centric Kinematics Analysis

“…Moreover, since the retroreflective markers block the depth measurements from the depth camera, the only way to recover the depth value for each marker is to use their surrounding information. To address the above problems, we proposed three algorithms: (1) Threshold analysis (Alg.1) − extending previous work in [25] to solve fast motion and camera noise during marker detection, (2) Marker detection (Alg.2) − the idea is to improve the marker centroid location accuracy and speed which are attached to joints of interest, in image space, (3) Depth recovery and mapping (Alg.3) − the 3D texture is partially missing in the marker region and it is possible to use the point cloud histograms for restoring the depth value of the marker centroid. When looking at the point cloud histograms, we can get a kernel that has higher weight inside according to their Euclidean distance to the marker centroid and frequency of occurrence.…”

“…There are several approaches to detect and identify blobs, such as matched filters / template matching [25], watershed detection [35], structure tensor analysis followed by hypothesis testing of gradient directions [36], [37], scale-space analysis [38]. All these approaches are limited by their sensitivity to noise, structure restriction and complexity [39].…”

“…All these approaches are limited by their sensitivity to noise, structure restriction and complexity [39]. In previous related work [25], a concentric cycle-based method (template matching) is proposed to perform the shape fitting test for each potential blob in order to locate all markers in image space (2D); however, this method is time consuming and requires expertise to determine associated parameters for the shape fitter and the kernel cluster filter, and cannot locate the center of the marker correctly when motion blur occurs and the marker is out of the sagittal plane, which leads to center deviation on those markers with circular distributed IR values.…”

“…2) Gait Phases Detection: Gait events of heel strike and toe off are used to measure the stance and swing duration using heel marker trajectory vertical axis values. An example is shown in Fig.7, where η 0 , η 1 , η 2 , η 3 denote inflection points, and ρ 0 , ρ 1 local extremum, D 2 via our proposed global gradient filtering algorithm, Alg.6, instead of using an averaging filter as in [25]. The proposed algorithm quantizes the heel marker vertical axis trajectory and then searches each quantization region between the inflection points from the global minimum to the maximum by iteratively regrouping the scanned points.…”

“…Evaluation of our proposed adaptive threshold analysis algorithm, Alg. 1, on an Intel i7-4710HQ 2.5GHz CPU, Win-dows10 OS, implemented by Visual C++, against the staticthreshold marker detection algorithm of [25], [42] indicates higher detection accuracy as shown in Table II. However, it introduces an extra preprocessing step which increases the processing time by roughly 1ms per frame.…”