Target Visual Tracking Technology Based On Correlation Filtering And Deep Learning

With the development of machine vision,visual target tracking technology now has a wider range of applications,and visual tracking for aerial targets has an unusual significance for airports,air defense and other fields.However,unlike other targets,air targets have the characteristics of fast moving speed,large deformation,and indistinguishable from the background.At the same time,objective factors such as weather and camera noise make it difficult to process and detect air targets.In view of the above problems,this paper mainly studies the detection and tracking of aerial targets,especially the target of drones,focusing on target detection algorithms and various visual tracking algorithms,and using Qt to develop interactive interface software to perform functional tests on the algorithms.Visualize the algorithm results in real time.The main works done are as follows:1.Study several commonly used image features and depth features to pave the way for the subsequent chapters.Then we study a method for static target detection: directly detect the air target using the deep learning-based YOLO detection framework.Through experiments,the method has good detecting results for static targets that have acquired prior information.2.Four classical visual single-target tracking algorithms are studied for aerial targets.Experiments show that these four methods can not meet the actual engineering needs for the latest application environment,and then the basic correlation filter tracking theory is studied.And two kinds of algorithms,through experiments,the DSST algorithm is the best algorithm for tracking the current target.3.Study several algorithms that have been introduced in recent years,which can be divided into the following three categories: correlation filtering based algorithms,correlation filtering algorithms using deep convolution features,and deep learning based algorithms.The application scenarios of these algorithms for aerial targets are tested.The overall tracking accuracy is high,but the required operating environment is also very high.4.The research of interactive interface development,using C++ and Matlab programming language,using Qt development framework to develop software interaction interface,various target detection methods,classic target tracking algorithm,correlation filtering based algorithm and its improved algorithm,using depth volume the correlation filter algorithm of the product feature are integrated into the software framework to provide various functions of the test algorithm,and display the tracking and detection results of the algorithm in real time.