Strengthen Aircraft Real-Time Multiple Object Tracking with Optical Flow and Histogram of Oriented Gradient Provided HSMA Implemented in Low-Cost Energy VPU for UAV

Abstract

Researchers and algorithm developers have always strived to make tracking more accurate than before, by developing deferents Algorithms that prove their efficiency and effectiveness in international competitions (VOT, MOT challenge, …), but taking care of one side without others may be the reason why limiting the fields of application of that software, which makes them confined only to these competitions or confined to applications domains without others and this is not the end object in itself. For example, the use of complex tracking algorithms in the drone makes the processing unit very slow and unsuitable for real-time tracking, instead that the use of highly efficient processing systems for acceleration consumes a lot of energy, which reduces the time the drone stays in flight. In this paper, we suggest our approach; HashSet Matching Algorithm (HSMA) provided by Optical Flow appearance features and HOG for developing an object-tracking algorithm to address the issue of a complex MOT algorithm for real-time tracking. The MOT challenge data-set experiment results show that the suggested technique can handle object occlusion difficulties successfully while reducing processing time. This algorithm is implemented in the Vision Processing Unit (VPU) to enhance the time flying of UAVs and to ensure low energy consumption. Our approach performs very well in real application scenarios, demonstrating that it is more practical.