Analysis of Walking and Running based on Markerless Model Ahmad Puad Ismail Faculty of Electrical Engineering Universiti Teknologi MARA 40450 Shah Alam, Selangor, MALAYSIA Email: puadismail@gmail.com Nooritawati Md Tahir Faculty of Electrical Engineering Universiti Teknologi MARA 40450 Shah Alam, Selangor, MALAYSIA Corresponding Author: nooritawati@ieee.org Abstract: This research investigated the possibility of side view human gait silhouette to be used for recognition of walking and running gait based on model-based approach. Markerless model with model based is used to produce the vertical angles of both hip and knee with respect to thigh for 32 image sequences as feature vectors for both legs for one complete cycle sequences. Overall, a total of 128 features are extracted based on four parameters from the lower limb of human body are validated for walking speed classification purpose. Further, the gait features extracted from different gait speeds is classified as walking and running gait using ANN and KNN. Initial findings with accuracy of almost 100% confirmed that the proposed method suited to be utilized as walking speed classification based on human gait. Keywords-markerless; model-based; Artificial Neural Network; K Nearest Neighbour; walking gait; running gait I. INTRODUCTION The utilization of video images to identify and analyze an event proven more frequently used recently. Although it is not promising hundred percent accurate results, but it might be helpful in identification of an object or human movement tasks. However the detection of objects is often not accurate when the recorded video failed to capture the human facial images. Therefore, an analysis can be performed to identify the subject through gestures and movement. Among the commonly used human movement is gait recognition where each human individual has a different and unique way of walking. Therefore, the special features were observed and analyzed to determine the sex, age and other characteristics associated with humans. Generally, to obtain the necessary data features for biometric process, model and non model based method is used. Model based methods require human model for each movement of subjects in a cycle while non model based method work only using the features of the image itself as the movement or shape analysis of images. Therefore, in measuring an effectiveness of feature extraction method and gait characteristic classification, ANN and machine learning are widely used as in [14], [5], [6] and [7]. II. LITERATURE REVIEWS Gait can be analysed and measured either using marker or markerless method. Further, human gait analysed using markerless modeling technique can be captured using model based or appearance based approach. Model based approach utilised modeling of body movement during walking at every frame of the gait sequences captured whilst appearance based employed image sequences as features are extracted for instance based on motion or shape analysis method [8] [9] & [10]. Additionally, model-based approaches incorporate knowledge of the shape and dynamics of human gait into the feature extraction process, constraining the expected shape and motion of the subject to a known set of possible alternatives that lead to overcome weaknesses of the previous category [12]. In this approach, the body model is fitted to the human in every frame of the walking sequence, and kinematic parameters are generally measured on the body model as the model deformed over the walking sequence [14]. Model-based approach normally will deal with the complexity of modelling and feature extraction but tend to be more attractive and promising. Further, human motion can be captured using marker and markerless method. Markers are utilized since precise motion information can be attained. Nevertheless, markers required intrusive specialized hardware and subject contact. On the other hand, markerless method captured motion of subject that is analysed for tracking and extracting objects, though not for biomechanical or recognition purposes [15]. For instance, Leung and Yang [13] proposed a new approach that automated markerless system for describing, analyzing, and classifying human gait by computer vision technique without subject contact or intervention and contributed as the an economical method that utilized silhouettes database of human image sequences. Research on human motion analysis related on markerless model has been conducted by other researchers that include incorporating the GLT-based motion estimation method in performing markerless human motion analysis [1]. As presented by K.Ogawara, X.Li and K.Ikeuchi in [2], whole body motion estimation method by fitting a deformable articulated model of the human body into the 3D reconstructed volume obtained from multiple video streams discussed. In addition, multi-view video sequences was used as an approach for modeling the human body by Sums of spatial Gaussians (SoG), allowing us to perform fast and high-quality markerless motion capture as in [3]. An innovative formulation of the joint constraints that enhances 2013 Fifth International Conference on Computational Intelligence, Communication Systems and Networks 978-0-7695-5042-8/13 $26.00 © 2013 IEEE DOI 10.1109/CICSYN.2013.51 212