Research On Uav Visual Object Tracking Method Based On Siamese Network

With the development of artificial intelligence technology,various high and new technologies require smart cars and UAV(Unmanned Aerial Vehicle)platforms for landing.UAV has become a research hotspot in the military and civilian fields thanks to their high mobility and flexibility among the platforms.The further improvement of UAV performance requires the support of visual object tracking while it is difficult to improve the current UAV visual object tracking algorithm.Due to the frequent interference of factors such as illumination,complex background,and deformation,it is difficult to maintain stable high-precision tracking for a long time in the UAV visual object tracking.Aiming at the current problems of deformation,occlusion and beyond field of UAV visual object tracking,this subject designs a visual network tracking algorithm based on siamese network to complete the long-term tracking for cars and persons.This paper does the research from following aspects:First,according to the complexity of UAV object tracking tasks,a multi-channel siamese network tracking algorithm is built as to increase the anti-jamming capability of the system,and the open source data set is used for training performance testing.Second,in order to solve the problem of deformation and partial occlusion in the process of UAV object tracking,this subject uses the cosine similarity algorithm to measure the similarity of pictures,and builds a template library with different convolution features.The verification in the aerial photography data set shows that the algorithm can effectively reduce the tracking interference caused by the deformation.Third,regarding the problem of full occlusion and out-of-field in the process of UAV object tracking,this paper proposes a new method for judging full occlusion and out-of-field,and puts forward new methods for local and global search algorithm.The verification in the aerial photography data set shows that the methods can effectively solve the problems beyond the field of view and increase the robustness of the system.Last but not least,this paper carries on the verification tests on multiple UAV aerial photography data sets,compares and analyses the multiple classic object tracking algorithms.The results indicates that the algorithm in this paper has higher average accuracy and average success rate.At the same time,the UAV object tracking simulation system under the ROS environment is built for testing.The results suggest that the tracking algorithm constructed in this paper can track normally under the ROS environment,with good stability and good practical value.