Kinesiology-inspired estimation of pedestrian walk direction for smart surveillance

Abstract

The article proposes a novel approach for estimating the walk direction of a pedestrian conjointly exploiting the complementary nature of video-based and frame-based walk direction estimation methods. Video-based direction estimation classifies inter-frame features by HMM as underlying machine learning method. To exploit the intra-frame features for orientation estimation, kinesiology based study has been performed and lower body kinematics of sagittal plane is identified to be the most suitable choice for feature. This frame-based intra-frame feature is classified using Least Square Support Vector Machine (LS-SVM) for finding orientation per frame. The score-level fusion of estimated results from both the topologies using a rule based look-up table produces the ultimate estimate for walk direction. The proposed method is tested over three different databases and is found to perform better than state-of-the-art methods due to inheritance of both inter-frame and intra-frame feature.