Tracking-by-Detection Optimization by IoU correction

Abstract

Real-Time Object Detection is a rapidly developing field nowadays but as a rule evaluating modern algorithms requires huge computing capabilities to achieve desired effect. This could be a main problem for these algorithms to be integrated into real world solutions. Evaluating of Visual Tracking Algorithms is the one of the most common solutions to this problem, but these algorithms show the high level of inconsistency. In this research I propose the new way for Visual Object Tracking Algorithms evaluation. While evaluating new method, it is expected to achieve increasing in Object Tracking stability yet preserving the positive improvement in evaluation speed. As a proof-of-concept, an application will be developed to show an effect of this approach with the ability to test it in the different cases.