Depth Energy Image for Gait Symmetry Quantification Caroline Rougier, Edouard Auvinet, Jean Meunier, Max Mignotte and Jacques A. de Guise Abstract— This paper introduces a new quantification method for gait symmetry based on depth information acquired from a structured light system. First, the new concept of Depth Energy Image is introduced to better visualize gait asymmetry problems. Then a simple index is computed from this map to quantify motion symmetry. Results are presented for six subjects with and without gait problems. Since the method is markerless and cheap, it could be a very promising solution in the future for gait clinics. I. INTRODUCTION Gait analysis systems are important for helping diagnostic of abnormal gait patterns. For simplicity, gait symmetry has been often used to characterize gait problems [5]. Indeed, the lower limbs are supposed to evolve symmetrically for a normal walker. This statement is controversial for some researchers as the gait can be influenced for example by limb dominance [6]. However, a quantification tool for gait symmetry could be useful for clinicians to evaluate walking dysfunctions, for example for stroke and amputee patients, or to analyze the recovery after a knee surgery. One commonly used method for gait analysis is motion capture (MOCAP) [8], [10] which consists in tracking in- frared (IR) reflective markers using multiple IR cameras. Such systems have been used to analyze gait symmetry [8], [10], as well as acceleration signals [9], with walkway systems [3] or laterally placed cameras [5]. In this paper, a new gait analysis system is proposed based on a treadmill associated with a cheap depth sensor placed at the back of the treadmill. The advantages of our system compared with MOCAP systems are that no markers are needed and its low cost price, which makes the system well adapted for clinical use. For our experiments, six young male adults were asked to walk on a treadmill (Life Fitness F3). After a period of habituation of 5min, their normal walk speeds were determined and used for further testing. Three tests were done: • Normal walk which served as a reference. • Right leg problem which was simulated with a heel cup (height of 2.5cm) placed inside the right shoe. This work was supported by the Fonds Qu´ eb´ ecois de la Recherche sur la Nature et les Technologies (FQRNT). C. Rougier, E. Auvinet, J. Meunier and M. Mignotte are with the D´ epartement d’Informatique et de Recherche Op´ erationnelle (DIRO), Universit´ e de Montr´ eal, Montr´ eal, Canada rougierc,auvinet,meunier,mignotte@iro.umontreal.ca J.A. de Guise is with the Laboratoire de Recherche en Imagerie et Orthop´ edie, Centre de recherche du Centre Hospitalier de lUniversit´ e de Montr´ eal (CRCHUM), Montr´ eal, Canada jacques.deguise@etsmtl.ca • Left leg problem which was simulated with a heel cup (height of 2.5cm) placed inside the left shoe. The heel cup is used here to generate a limping walk which will produce an unbalanced gait with asymmetric character- istics. For each test, after another period of habituation on the treadmill (2-3 min), a three-minute video was recorded with the depth camera (see Section II) placed at the back of the treadmill (back view of the person). Ethical approbation was obtained from the research ethics board (REB) of our university for this project. II. DEPTH SENSORS Depth maps, which show the different depths of a scene, can be obtained in several ways: • Stereo vision [13] The 3D view of a scene can be reconstructed with a calibrated binocular system. How- ever, to obtain precise depth maps, such systems require to be well calibrated and to have a textured scene. Moreover, stereo reconstruction algorithms are often computationally expensive. • Time-of-Flight (TOF) camera [14] Accurate depth images can be obtained with a TOF camera, but this technology is very expensive and currently limited to low image resolution (e.g. image size of 176x144 pixels in [7], [14]). • Structured light With a known artificial texture pro- jected on the scene, a depth map can be obtain from a monocular system. The Kinect sensor [11] is based on this method with an infrared structured light (IR dots) projected in the scene and observed with an infrared camera. Such systems can acquire bigger images than a TOF camera at a lower price (e.g. image size of 640x480 pixels at 30 fps for the Kinect sensor which is currently fifty times cheaper than a TOF camera). For clinical gait analysis, a low-cost and easy-to-install system is more suitable, which encouraged us to choose the Kinect sensor [11] to acquire depth images. The resulting images are disparity maps where far objects are represented with higher Kinect disparity values (within the depth range used in our study). The disparity values can be converted in depth values after a calibration step, which consists in moving a plane along a rail at known depths and acquiring corresponding disparity values. Then, a set of disparity-depth pairs is obtained and used to compute the relation between disparity and depth: Depth =1/(−0.0032936 Disparity +3.5463) (1) An attempt to use depth images for gait analysis has previously been done using a TOF camera [7]. However, 978-1-4244-4122-8/11/$26.00 ©2011 IEEE 5136 33rd Annual International Conference of the IEEE EMBS Boston, Massachusetts USA, August 30 - September 3, 2011