Behavior Analysis of Robot Operating System for Motion Tracking and Occlusion Handling A Social Robot Application

Abstract

Human detection, identification and tracking plays an important role in newlineapplications ranging from surveillance, robotics, entertainment and day newlinecarrying activities. The advancement in the technologies opened a new newlinewindow of automation where human-machine (i.e. robots) interaction found to newlinebe promising approach to increase the flexibility of the human work. This newlinethesis presents the motion tracking algorithm for the robotic applications. newlineEven though many algorithms have been proposed to track the motion of the newlineobjects, it is a challenging task due to variation in the human pose, complexity newlinein the background, illumination variation, occlusion etc. In addition, the robotic newlineapplications add new constraints of real-time and accurate tracking of human. newlineTherefore the basic goal of the thesis is to implement a human tracking newlinealgorithm for robotic applications. In initial phase, algorithm is validated for the newlinesurveillance dataset and later real-time analysis is presented in the thesis. newlineMotion segmentation is the first step to track the human. In this work, motion newlineis segmented using background subtraction algorithm which models every newlinepixel of the frame as a mixture of Gaussian distribution function and finds thelocation of the all the humans present in the frame. An algorithm is proposed newlineto examine the Gaussian distribution resulting accurate segmentation of the newlinehuman from every frame. The proposed evaluation model achieves 97% newlinehuman detection. Second challenge in the robot based tracking model is that newlineit has to retain the identity of all human under the occlusion during the tracking newlineprocess. Multiples inkling including size of bounding box, velocity etc can be newlineused to retail identity. newlineIn the proposed work, an algorithm using the statistical features looking newlineto the constraints of real time requirement is proposed. The proposed model newlineis capable to handle and retains the identities even under the long term newlineocclusion. Third constraint in the robotic application is the grow of robots newlinescale and scope. This leads to utilization