20th IMEKO TC4 International Symposium and 18th International Workshop on ADC Modelling and Testing Research on Electric and Electronic Measurement for the Economic Upturn Benevento, Italy, September 15-17, 2014 2D video-based human gait analysis: a novel markerless approach A. Castelli, G. Paolini, A. Cereatti, U. Della Croce Information Engineering Unit. Polcoming Department, University of Sassari Interuniversity Centre of Bioengineering of the Human Neuromuscoloskeletal System V.le Mancini 5, 07100, Sassari, Italy. e-mails: acastelli@uniss.it; gpaolini@uniss.it; acereatti@uniss.it; dellacro@uniss.it Abstract – A 2-D markerless technique for the analysis of the lower limbs kinematics during gait is presented and preliminarily validated. Data were recorded using a single video camera placed laterally to the walkway. The preliminary validation was performed using a optoelectronic marker based system as gold standard. The lower limb facing the camera was modelled with four body segments: foot, shank, thigh and pelvis. Ankle socks and underwear garments were used as segmental markers to track the foot and the pelvis, respectively. The shank and thigh were tracked using multiple reference points defined in a calibration reference image. The comparison between the estimated segment angular kinematics and that obtained with the gold standard revealed a high correlation index and limited Root Mean Square Deviation (RMSD) values. The proposed technique can be considered as an easy-to-configure and affordable alternative to marker-based systems, for 2D human lower limb motion tracking during gait. I. INTRODUCTION Quantitative human gait analysis is often used in clinical environments. Passive marker optoelectronic systems are widely used to this purpose. Although accurate and reliable, they are expensive and require space and technical expertise, therefore their use is limited to few facilities [1]. To increase the use of quantitative gait analysis in clinics, including ambulatory environments, low cost, video-based markerless (ML) techniques have been seen as a potential valuable alternative [2]. Moreover, as an additional advantage, the use of a ML technique is expected not to affect the function being measured, since it does not require any fixture applied to the skin of the patients [3]. Finally, ML solutions requiring a single color camera could be applied when a 2D sagittal joint kinematics analysis is sufficient, with a consequent reduction of the setup time, complexity and cost. In a previous study, we proposed a 2-D ML technique for the estimation of ankle kinematics [4]. In this study, we present a model-based ML method for the analysis of gait with a single color camera positioned laterally to the walkway. The ML method estimates the sagittal angular kinematics of the pelvis, thigh, shank and foot of the limb facing the camera, solving the lower limbs superimposition occurring during both the swing and stance phases of gait. A preliminary validation of the technique is carried out, by comparing the segmental angular kinematics estimated with the ML method to that obtained with simultaneous measurements provided by a stereo-photogrammetric system (gold standard). II. MATERIAL AND METHODS A. Experimental protocol Nine healthy subjects (8 males and 1 female, age 33 ± 6 y.o.) wearing white ankle socks and white underwear garments, were asked to walk along a walkway. Subjects walked at three self selected speeds (comfortable, slow and fast), to verify if gait changes related to velocity would affect the ability of the method to cope with lower limb superimposition. Five trials per selected speed were captured. A RGB video camera (Vicon Bonita Video, 1280x720p, 50fps) was placed laterally to the walkway. Lens distortion was corrected using the Heikkilä undistortion algorithm [5]. A homogenous blue background was positioned on the opposite side of the walkway. A static reference image was captured at the beginning of the data acquisition session. Video files were processed using MATLAB (2010b, The MathWorks, Natick, MA). For validation purposes, marker-based data, synchronous with the video data, was captured using a 6- camera stereo-photogrammetric system (Vicon T20). The retro-reflective markers were placed as follows: two markers on the lateral surface of the underwear garment, one marker on the Greater Trochanter (GT), one on the femoral Lateral Epicondyle (LE), one on the Lateral Malleolus (LM) and two were placed on the lateral surface of the sock (Fig. 1a). The marker set used allowed to define the same segment reference systems for both the ML and marker-based techniques. B. Model Calibration Rigid templates of the lower limb body segments were defined from the static reference image. The operator identified LM, LE and GT by means of a mouse click on ISBN-14: 978-92-990073-2-7 20th IMEKO TC4 International Symposium and 18th International Workshop on ADC Modelling and Testing Research on Electric and Electronic Measurement for the Economic Upturn Benevento, Italy, September 15-17, 2014 760