Research On Vision Tracking Of A Mecanum Wheeled Mobile Vehicle With Intermittent Image Losses

Omnidirectional mobile robot has gained popularity because its ability of turning in any direction without reorientation and capability to move in confined spaces.On the other hand,to improve the autonomy of mobile robot,a widespread use of vision systems has been introduced to robotics applications and tracking a moving target is an important one.However,vision systems suffer from the problem of intermittent image losses,usually caused by occlusion,feature matching failure,etc.This thesis is concerned with the target tracking problem for a Mecanum wheeled mobile robot(MWMR)with camera in the presence of intermittent image losses.The robot may experience image losses for many times during tracking.The concept of switching system is used to model the problem.Then series work including design of switching control scheme,stability analysis,introducing of stochastic model is conducted.The main contents are developed as follows.Firstly,a switching model is used to describe the tracking problem.The model is built under the MWMR body frame.In this way the onboard sensors including camera can be used and described.The state of the system is defined as position and orientation error between the MWMR and the target.The control inputs of the system are defined as 3-Do F velocities of the MWMR.Due to the intermittent image losses,the model can switch between two modes: one for the case when the target image is acquired,and the other for the case when it is not.Traditional methods to analyze switching systems and the kinematics of a MWMR is discussed to give the following content a foundation.Secondly,a switching control scheme and a stability criterion are given.Two control strategies are designed corresponding to two modes.When the image is acquired,the controller is designed to make the close-loop system exponentially stable.When the image is lost,a predictor is given to produce the estimated position and orientation of the target.The controller is then designed to allow the tracking of the estimated target to be exponentially stable.The gains of the controllers are switched to suit different modes.Challenge emerges when analyzing stability of the tracking system.During the mode of image lost,the tracking error may increase when the actual motion of the target is quite different from that produced by a predictor,and also the tracking error is not controllable totally,making the existing methods hardly to be applied directly.To go over this problem,a regulargraphic containing possible target position is developed.By finding the largest distance between MWMR and points in the region,the worst case of tracking can be considered,therefore gives the tracking error an upper bound.In this way the stability criterion for the tracking system can be acquired.The criterion considers about the relation between duration times of two modes.Simulation platform is established.Simulation results verify the effectiveness of the proposed methods.Finally,the problem of stochastic image losses is investigated.Stochastic model is introduced by modifying the switching system to a continuous-time Markov jump system.By introducing the stochastic model,more information about the duration times of the two modes can be given.Specifically,the duration times satisfy exponential distribution.By focusing on the increase and decrease in one cycle,the expectations of the variation of the tracking error during two modes are derived.To make the total trend of the expectation of the tracking error decrease and converge,the relation between controller gains of the two modes is given.Under these controllers,the system is proved to be mean square stable.