Multiple Moving Object Video Visual Tracking Based On Part Graph Decomposition

With the development of intelligent equipment,people can get a great quantity of information which of the most is intuitive visual information,which makes computer vision has become one of the important research content in many scientific research fields.Object tracking is a challenging problem and also an important task within the field of computer vision.It is widely applied for motion analysis,video surveillance,robot navigation and human-computer interaction.Among these applications,the object tracking algorithm is required to be robust and effective.Thus,the multiple object tracking task is a very challenging research topic.Therefore,it is of great significance and value to design an effective multi-object tracking algorithm.With the continuous development of computer processing and enhancement of the performance,people put forward a series of multi-object tracking algorithms,and these algorithms have advantages and disadvantages.However,it still cannot adapt to the multiple object tracking applications in complex scenes.In the real complex scenes,there will be the illumination change,targets crossing or target occlusion between each other,which makes the scenes become complex and changeable in video sequence.Therefore,multiple object tracking at same time could be very challenging.In this thesis,the existing multiple object tracking algorithm is studied,and we propose an algorithm that combines various complex background information,in the process of tracking the use of sub graph decomposition algorithm for multiple targets tracking.The research work of this thesis is divided into two parts:First of all,in order to establish a stable and effective object in real environment and expression method,this dissertation analyzes the characteristics of light motion information,blurred information,put forward the target under complex background appearance expression modeling method.In this dissertation the appearance expression model combined with the depth information,illumination information and movement to achieve the target on the expression of blur information using multilayer framework,can adapt to the appearance of object deformation,motion blur,illumination etc.The experimental results show that the object tracking algorithm based on the target expression model can reduce the impact of cluttered background and environmental illumination changes,and the average success rate of target tracking is improved by more than 2%.Secondly,according to the interaction between multiple objects occlusion,we will use the subgraph decomposition strategy in the process of multiple objects tracking,through the introduction of disjoint paths and the connected probability model,in order to process subgraph decomposition,effectively a graph partitioning problem is transformed into a minimum cost subset decomposition problem.In addition,a variety of strategies according to the algorithm,the problem of tracking object segmentation is more detailed and is able to track the scene too crowded or complex,or between the objects occlusion case,can track the target more stably,enhances the tracking accuracy standards,improves the robustness of the algorithm.This part of the work has been collated and published in the International Conference on IEEE ICCE 2016.In this thesis,testing sequences are released from Multiple Object Tracking(MOT)Challenge 2015 video collection which is a challenge project in IEEE CS.In this thesis,a multi-object tracking algorithm is proposed in this thesis,and the results are compared with the experimental results.In the experiment,the multi object tracking algorithm proposed in this paper has good performance,and the average tracking success rate is improved by 2.8% compared with the adaptive light,blurred and object occlusion.