Object Tracking Algorithm Based On Deep Learning And Scale Adaptive Correlation Filter

Object tracking is an important and challenging work in the field of computer vision,which has been widely used in areas such as unmanned driving and security monitoring.In recent years,object tracking develops very quickly.However,due to the complexity of realistic environments,there are still many problems to be solved in the process of object tracking.Firstly,based on the comparison of different methods of feature extraction and in order to solve problems such as tracking drift in object tracking,the VGG-19 deep convolutional neural network model is used for feature extraction and a multi-layer convolutional feature fusion mechanism is applied,which promote effective and robust deep feature representation.Secondly,in order to mitigate the time-consuming problem caused by the large amount of deep learning calculations,this thesis is researched to a kernelized correlation filters object tracking algorithm that can achieve fast calculations.But this algorithm has weak adaptability to target scale changing.To solve this problem,two strategies for scale adaptation are introduced to enhance the adaptive ability of algorithm while the target scale changes,which can better estimate the target.Then,combining above improved model for feature extraction and improved framework of kernelized correlation filters,this thesis gives two visual object tracking algorithms and extensive experiments are performed on the public dataset OTB2015.Compared with the mainstream tracking algorithms,the results show that both the proposed algorithms have better average success rate and average precision rate,and show better coping ability in complex environments such as scale change and fast motion.Finally,the optimal algorithm implemented in this thesis is applied to object tracking videos taken by drones and the results are analyzed.In addition,an object tracking algorithm learning software system is designed to help beginners in the field of computer vision to better understand various visual object tracking algorithms.