Calculating Body Segment Inertia Parameters from a Single Rapid Scan Using the Microsoft Kinect

Abstract

Many biomechanical analyses rely on the availability of reliable body segment inertia parameter (BSIP) estimates. Current processes to obtain these estimates involve many time consuming manual measurements of the human body, used in conjunction with models or equations. While such methods have become the accepted standard they contain many inherent errors arising from manual measurement and significant assumptions made in the underlying data used to form the models and equations. Presented here is an alternative approach to obtaining reliable estimates of body segment inertia parameters through the use of the Microsoft Kinect sensor. A 3D scanning system was developed, comprising four Kinects aligned to a single global coordinate system using rigid body calibration and random sample consensus (RANSAC) optimisation. The system offers the advantage of obtaining BSIP estimates in a single scanning operation of around three seconds, much quicker than the circa thirty minutes of manual measurements required for existing BSIP estimation methods. The results obtained with the system show a mean error of 0.04% and a standard deviation of 2.11% in volumetric measurements of a torso manikin, suggesting comparable and in many cases, greater accuracy volumetric estimates than a commonly used geometric BSIP model. Further work is needed to extend this study to include a full range of BSIP measurements across more of the bodies segments and to include scanning of living human subjects. However, this initial study suggests great potential for a low cost system that can provide quick and accurate subject specific BSIP estimates.

Assessing the Suitability of the Microsoft Kinect for Calculating Person Specific Body Segment Parameters

“…The Microsoft Kinect NUI sensor (Microsoft Corporation, Redmond, USA) was chosen for this study, owing to its low price point, support of a full software development kit (SDK), and the favourable results presented in previous studies [37][38][39].…”

“…A single computer running custom software (created using the Microsoft Kinect SDK-Microsoft Corporation, Redmond, USA) was used to control the Kinects, perform calibration, and capture scans. The software was used to switch the Kinect's infra-red (IR) projectors on and off during scanning to prevent interference between neighbouring sensors [38]. This resulted in a scan time of 1.5 seconds, simultaneously capturing 3D and colour data.…”

“…The local coordinate system of each Kinect was aligned to a global frame using an initial calibration procedure [38]. A calibration object -comprising four spheres mounted on a vertical rod (figure 2a) -was placed in nine different positions within the capture volume.…”

“…Sphere centres were identified using a combination of Hough transforms [43] and a minimisation technique (figure 2c and 2e). A rigid body transformation algorithm [41] and RANSAC optimisation [38] used the sphere centre locations to calculate the requisite transformation matrices. …”

“…A number of recent studies [31,36,37] investigated the accuracy of typical NUI sensors, but mainly focussed on simple measurements such as Euclidian distances and plane fitting residuals. Only single sensors were investigated, whereas a body scanning system would typically comprise multiple sensors [30,38,39], possibly leading to a computing of error. A recent study by Clarkson et al [39] provides the most applicable assessment, using a scanning system comprising four Microsoft Kinect NUI sensors to take multiple scans of a machined cylinder representative of a large body segment.…”