Object Tracking Research Of Moving Robot Based On Multiple Instance Learning

With the continuous development and extensive application of artificial intelligence,it has already risen to the national strategic level.Robotics as an integrator of artificial intelligence technology,and it is drawing more and more researchers' attention on the practical application.It is also a core problem needed to be solved that the autonomous object recognition and tracking of the mobile robot based on the intelligent technology.How to migrate the improved machine learning algorithms to mobile robots,and makes it more robust to handle problems such as light changes,occlusion,and complex backgrounds,this is still a very challenging research technique.In this paper,based on the wheeled mobile robot platform MT-R,we achieved the autonomous target tracking and control through the improved machine learning algorithm,so as to intelligent application can be finished further for the wheeled mobile robot.The main works of this paper can be summarized as:Firstly,we review the research status of visual object tracking,and list the methods used in different visual object tracking algorithms as well as the shortcomings of these algorithms.This paper focuses on the analysis of the tracking method based on multiple instance learning and the co-training algorithm.What's more,it has become the theory foundation of the proposed object tracking method.At the same time,the research and development of mobile robot on the world are briefly discussed.Then,the internal object tracking algorithm of mobile robot introduced in detail.Tracking-by-detection tracking methods have achieved favorable performance in recent years.Nonetheless,most of these approaches employ sample collection as a task that is independent of classifier training and ignore the correlation between them.This often leads to the unstable performance of classifiers due to the lack of discriminative samples.In this paper,we present an improved multiple instance learning algorithm which prevents model drift significantly.Secondly,in order to improve the capability of classifiers,an active sample selection strategy is proposed by optimizing a bag Fisher information function instead of the bag likelihood function,which dynamically chooses most discriminative samples for classifier training.Thirdly,we integrate the co-training criterion into algorithm to update the appearance model accurately and avoid error accumulation.The tracking results demonstrate the strength of our approach by a large margin on multiple publicly available datasets including the challenging sequences.Finally,the proposed object tracking algorithm is running on the mobile robot MT-R,combing with the robot hardware motion-driven strategy,finished the mobile robot's autonomous object tracking and moving.The experiment results under different actual scenarios verify the robustness of mobile robot object tracking system,the proposed object tracking algorithm can effectively help mobile robots handle the problem such as partial occlusion,illumination changes etc.