Research On High Speed Visual Object Tracking Algorithms

In recent years,with the popularization of computers and the rapid development of hardware technology,the information discipline has begun to step into the era of big data.Artificial Intelligence(AI),Machine Learning(ML),and Deep Learning(DL)are powerful tools to deal with big data and have attracted most interests of researchers.Although these disciplines are burgeoning,it is still difficult to keep up with the explosive growth of data.The mathematical algorithm as a core part of these may be a key factor to break this bottleneck.Visual object tracking combines the knowledge of AI,ML,DL and Image Processing and has a wide range of significant applications in the real scene.How to automatically and intelligently implement accurate analysis of video sequences in real-time is an extremely challenging task.The correlation filter-based tracking algorithms have become the most popular framework in the field of object tracking because of its high speed of tracking and in this paper we mainly concern about them.Many correlation filter-based tracking algorithms have achieved good tracking performance.The spatio-temporal context(STC)object tracking algorithm is a correlation filter-based algorithm that has a very fast tracking speed.However,the tracking accuracy is very limited due to the use of grayscale features and difficult to be applied in the real scene.Focusing on the issue that the tracking accuracy of STC algorithm is insufficient,a new method of applying improved Histograms of Oriented Gradients(HOG)features to STC tracking algorithm is proposed in this paper.We improved the algorithm from the following two aspects.First,we choose to use the improved HOG features.Compared with the original HOG features,the improved HOG features are more capable of representing the target,while the dimensions are reduced.Second,we introduced the kernel methods to overcome the curse of dimensionality and the Gauss kernel function is used to perform the multi-channel features fusion.Based on this,the STC model working on the multi-channel HOG features is established.Besides,the fast Fourier transform is used to accelerate the computation,which ensures a high tracking speed.A large number of experiments show that the improved algorithm has a significant increase in accuracy compared with the original STC algorithm.In addition,the improved algorithm has a better performance than Kernelized Correlation Filters algorithm both in accuracy and speed,which indicates that it is effective to apply the improved HOG features to STC visual object tracking.