The Research Of Cooperative Multi-robot Tracking Of Multiple Moving Targets Supported By Cloud Computing

Cooperative robot tracking of multiple targets is a classic problem in robotics field,and it plays an important role in a wide range of applications,such as security patrols,reconnaissance search and so on.In this issue,the task of tracking multiple moving targets is coordinated by several robots cooperating with each other to achieve.Multi-robots maximize tracking efficiency by collaborative planning and target switching when necessary.Under the cloud robotics architecture,cloud computing is expected to provide background support for cooperative robot tracking of multiple targets tasks in two aspects: On the one hand,the "cloud" can be used as the collaborative infrastructure of groups to bring together high-level cognitive views at the group level,and then use the coordination of the cloud planning ability to dynamically achieve more efficient tracking task distribution.On the other hand,the robot can accumulate a great deal of knowledge of the target(such as the characteristics in different lighting conditions)when tracking a specific target.The existence of cloud makes it possible to reuse this knowledge among different robots,thus significantly improving the tracking accuracy.Based on the architecture of the cloud robot,guided by the above ideas,we focus on the design and implementation of cooperative robot tracking of multiple targets system supported by cloud computing.Focusing on the system architecture,key technologies and prototype implementation,the following three aspects are studied:(1)Proposed the architecture supported by cloud computingWe proposed the cooperative robot tracking of multiple targets architecture of "back-end cloud + front-end multi-robot".In this architecture,the back-end cloud aggregates the status and target knowledge of multiple robots from the front to form a community view,so as to provide task assignment support multiple robots target tracking.At the same time,the tracking engine was placed in the cloud,gaining knowledge through on-line fine-tuning,and knowledge sharing via on-demand engine switching of engines.(2)Designed a task allocation and knowledge sharing mechanism that supports multi-target trackingUnder the above architecture,we design the task allocation and knowledge sharing mechanism.In the aspect of task assignment,the relationship between multiple robots and multiple targets is maintained through the acquisition and updating of group views,and the back-end method is used to dynamically allocate tasks in the cloud based on the views.In terms of knowledge sharing,the on-demand switching of engines allows multiple robots to share the knowledge accumulated by the cloud-tracking engine.In addition,aiming at the inconsistent state caused by communication delays between the cloud and the robot,a frame recognition mechanism is introduced in this topic.(3)Constructed cooperative robot tracking of multiple targets prototype system supported by cloud computingBased on the breakthrough of the above key technologies,we build a prototype system that supports cooperative robot tracking of multiple targets.Compared with three classical distributed mission planning algorithms,we comparatively analyzed the improvement of the average observation rate of the target based on the group view mission planning through simulation.In addition,based on the open data OTB50 and VOT2014,experiments were carried out to improve the tracking accuracy by knowledge sharing.Finally,the prototype system was verified in a real environment.