Research On Object Tracking Methods Based On Deep Learning

In the past few decades,computer vision and multimedia understanding have become important research areas in computer science.At the same time,visual tracking of moving targets is one of the most important applications in computer vision,and has become a hot spot in current research.Although much progress has been made so far,object tracking is still widely recognized as a very challenging task due to many reasons such as occlusion and illumination.With the rise of the third wave of artificial intelligence centered on deep learning,deep learning has broken through a series of challenges in computer vision and natural language processing with its powerful problem-solving capabilities.With the continuous growth of data scale,deep learning has achieved significant results in target detection and image classification.In recent years,object tracking algorithms based on deep learning have also been gradually proposed,which has made a major breakthrough in the field of object tracking.This paper focuses on object tracking algorithms based on deep learning.First of all,this paper reviews the state-of-the-art object tracking algorithms that have been applied deep learning in recent years,mainly based on CNN features and correlation filtering or end-to-end neural network.Then,this paper takes SiamFC algorithm as the baseline,and improves it from tracking strategy and network structure.In terms of tracking strategy,this paper introduces the re-detection mechanism,and improves the discriminative ability of the algorithm by using the generated model to construct template and the high-confidence model update strategy.In terms of network structure,by using the deeper ResNet18 instead of AlexNet,and incorporating multiple layers of features,the target can be positioned more accurately.Tests on public datasets illustrate the superior performance of our improved algorithms.