Research On Object Tracking Algorithm Based On Deep Learning

Object tracking is one of the most important and difficult problems in computer vision.The application and development of target tracking is of great significance,for example,it plays an important role in the both military and civil fields such as unmanned aerial vehicles(UAV),intelligent transportation,precision guidance,robotics,human-computer interaction and so on.Target tracking has been studied for several decades,and many achievements have been achieved.However,although many tracking algorithms have been proposed,it is still a great challenge to realize robust tracking due to the complexity of the tracking process.With the rise of machine learning algorithm,deep learning is used to solve the problem of computer vision and other fields.So deep learning is naturally introduced into the target tracking problem.Under this circumstance,the tracking can be regarded as a two classification problem and will be separated the target from the background.Deep learning is a multi-layer neural network to simulate the process of human's brain learning and analysis.This structure can better learn the essential characteristics of the object.In this paper,a deep neural network model with strong feature extraction capability is used to track the target from the perspective of feature extraction.The aim of this paper is to explore the effect of target tracking algorithm based on deep learning.Aiming at the problem that the traditional tracking algorithm is easy to be disturbed by the noise in the complex environment,stacked denoising autoencoder(SDAE)is used as the network structure of the deep model in this paper.By using the data contaminated by noise in the pre-training,the anti-jamming ability of the algorithm is enhanced.Aiming at the problem of poor classification of logistic regression classifier in traditional deep network,support vector machine(SVM)is introduced at the top of the network,which enhances the classification ability of the network and improves the accuracy of the tracking algorithm.In the process of tracking,this paper uses the particle filter framework to complete tracking.The particles were dispersed by the particle filter algorithm,and all particles were passed through the deep network.The confidence of each particle is obtainedthrough the deep network,and the particle region of the highest confidence is taken as the final tracking result.Finally,in order to test the performance of the proposed algorithm,the image sequences of various complex environments are tested on the VTB standard test platform.The experimental results show that the proposed algorithm can effectively solve the problems such as occlusion,illumination variation,scale variation,deformation and background clutters.