Research On Automatic Visual Surveillance Using Multiple Mobile Robots Based On Distributed Consesnsus Algorithm

Visual surveillance has been one of the most active research areas in robotics and computer vision.Its main purpose is to use a series of images obtained in the scene to detect and track specific targets.Visual surveillance system has a wide range of applications,such as traffic surveillance,security guarding of communities and buildings,detection of military targets,medical monitoring system,etc.Traditional visual surveillance systems usually deploy several cameras in the scene,transmitting the video data to the central monitoring room,where a series of images are displayed to security personnels.The main problem of traditional surveillance system lies within the nature of the security personnel's missions.People are required to passively watch a number of screens and may lose their focus over time.Besides,human beings are not able to notice every small detail of a picture,let alone capture details of multiple targets simultaneously.An efficient scheme is using robots equipped with cameras to automate this process.In this dissertation,firstly feature detection and image matching techniques are utilized to extract common targets among different perspectives.To meet the reliability and real-time requirements of multi-robot system,we then compare different algorithms and adopt the ORB algorithm to perform matching tasks in our work.Besides,we utilized the K-Nearest Neighbor algorithm to improve the accuracy of the ORB algorithm.Based on the matching results obtained,we introduce the consensus protocols of multi-agent system to design the control algorithm.Here the interaction topology is mainly considered to be dynamically changing.In this dissertation,the automatic visual surveillance problem is formulated as a visual consensus problem and propose algorithms to control all the robots to automatically adjust its poses and track the common target.Since the robots may lose the views of target during the motion,we utilize the image-based visual servo control strategy to ensure that the mobile robots always meet the visual visibility constraints.We briefly analyze the stability of the proposed algorithms based on graph theory and stability theory.Besides,the feasibility and effectiveness of proposed algorithms are demonstrated respectively by simulations implemented on the Virtual Robot Experimentation Platform(VREP)and experiments tested on the Pioneer-3DX robot platform.Finally,the main work and contribution of this dissertation are summarized.Some shortcomings are also referred and we make some prospects for the follow-up research work.