Real-time deep learning based visual object tracking

Abstract

With the speed of advancement in artificial intelligence technology nowadays, itis not naïveto image the unimaginable.The people livingin the pastwould not have fandom to be able to track people by surveillance cameras.The first form of artificial intelligence appears in the late 1940s, which gradually branchesout to machine learning and deep learning today. There is agrowing interest on deep learning in recent years, andthe topic is very popularin the researchfield currently. The popularity of this topic can be noticed asthere isan increase in numbers of research papers publishedregarding deep learning. Deep learning’simpact in theindustry startedin the early 2000sbut its’big scaleimpact on the various industrial application began roughly in 2010. Thus, there is still plenty to be researched on and improved. An example ofsuchapplications of deep learning would be visual object tracking.A challenge faced byvisual object tracking would be occlusion, a very common problem facedin image processing. Then, there is also the problem of model drifting, when unforeseencircumstances appearoutside of what the model is trying to estimate, changes over time.