Research Of Vision-based Ground-Air Robots Cooperation Methods

Since the advantages and disadvantages of ground mobile robot and UAV can complement each other, they can complete the tasks by collaboration which reflects the remarkable excellence of the system. The ground-air robots cooperation system plays a more and more important role in many military and civil fields, and becomes the research hotspot of many domestic and foreign research institute. Using the ability of UAV that can monitor a wide range environment to assist ground mobile robot navigation is the main working mode of the ground-air robots cooperation system. UAV has wide field of vision which can be used for rapid ground target location, but it is not as accurate as the ground mobile robot for ground tasks. So the ground-air robots system make up for each other's shortcomings, to complete their tasks through coordination. In this paper, on the basis of stable tracking of the ground moving robot, taking the collaborative obstacle avoidance system as the main research direction, the main method of the ground-air coordination control system is studied. The main work of this paper is as follows:Firstly, the HOG feature is selected as aircraft detection and the main visual feature of the ground environment. The shape and contour of the local area can be well described by this feature, so it is widely used to detect the rigid object. In summary, the HOG feather is used to detect and locate the target of the ground mobile robot. However. due to the character of the gradient. HOG feature is difficult to deal with problem of image edge under the complex environment. In order to solve such problems, this paper gives an improved method of the feature extraction.There are many excellent tracking algorithm in academic fields. and KCF tracking algorithm is one of them. The advantage of KCF lies in the use of cyclic shift dense sampling methods and classifiers training with fast Fourier transform, which make its tracking speed be outstanding. However, when the target is under partially and fully occlusion, the KCF tracker update the classifier and appearance model with the disturbed samples, leading to the eventual failure. This paper proposes a robust occlusion detection scheme based on this method with complement of the KCF tracker and Kalman filter. When the target is under partially and fully occlusion, predicting the state information by Kalman filter and stopping updating the classifier and appearance model, thereby which improves the ability of the tracker in processing the occlusion. On this basis, the proposed algorithm simulation experiments verify the feasibility and effectiveness of this method.Then, the components of the cooperative control system are introduced in detail, which includes the introduction of the hardware of UAV and the mobile robot. For the design of the collaboration control platform, the communication of UAV and ground mobile robot and the preprocessing of collected image are introduced in detail. At last, the concrete workflow of the cooperative control system is introduced.Finally, the experiment achieves the stable tracking of ground moving robot with the proposed tracking algorithm on the platform of designed coordinated control system. And the ground-air cooperative obstacle avoidance strategy is given.Taking UAV and the ground mobile robot as the control object,a series of experiment for the proposed tracking algorithm and cooperative obstacle avoidance strategy in the artificial experimental environment is carried out. Various interference factors are processed in those experiments and the results show that the proposed scheme is feasible and effective.