Research On Visual Object Tracking For Unmanned Aerial Vehicle

Nowadays,UAVs have been widely used in various fields,visual object tracking algorithms is very important for the application of UAVs.Firstly,Under the unmanned aerial vehicle platform,the visual object tracking has high requirements on the tracking algorithm,and it is required to detect the tracking object accurately under the complex scene and the realtime condition.Aiming at the problem that the traditional kernelized correlation filter tracking algorithm lacks scale estimation and weak ability against occlusion,a scale adaptable multi-feature fusion object tracking algorithm based on kernelized correlation filter is proposed.The algorithm used the position estimation module to get the object location,and then utilized the scale estimation module based on multi-feature fusion to get the scale estimation result.Then the small sample bank was constructed and the sample updating strategy has been adopted.Experimental results from the public dataset and unmanned aerial vehicle dataset show that the algorithm can quickly and accurately track the object under the premise of satisfying the real-time requirement,and it also has robust performance under the complex occlusion and scale variant.Secondly,the scale adaptable multi-feature fusion object tracking algorithm based on kernelized correlation filter is not sensitive to the target deformation,so we propose a hybrid long-term object tracking algorithm based on convolutional neural network.The experimental results show that the algorithm can effectively solve the problem,and the whole performance of the algorithm is excellent,it also can fit the application requirements when used on UAV platform.