In computer vision, object tracking remains a pivotal challenge, significantly impacting the performance of various applications such as autonomous driving, surveillance, and robotics. The Intersection-Over-Union algorithm is a lightweight and efficient method for online multi-object tracking. This characteristic allows it to be deployed on low-performance devices, including edge devices and battery-powered devices such as unmanned aerial vehicles. Unfortunately, the performance of the algorithm depends on several parameters that, if not appropriately sized, can make it useless. This paper deeply characterizes these dependencies by means of a Monte Carlo exploration of the design space. As a result of the empirical experiments, an optimal combination of input parameters is proposed to maximally increase the performance without affecting the lightness of the algorithm.
Tuning Intersection-Over-Union Algorithm to Enhance Tracking Performances
Calisti L.;Contoli C.;Kania N.;Lattanzi E
2024
Abstract
In computer vision, object tracking remains a pivotal challenge, significantly impacting the performance of various applications such as autonomous driving, surveillance, and robotics. The Intersection-Over-Union algorithm is a lightweight and efficient method for online multi-object tracking. This characteristic allows it to be deployed on low-performance devices, including edge devices and battery-powered devices such as unmanned aerial vehicles. Unfortunately, the performance of the algorithm depends on several parameters that, if not appropriately sized, can make it useless. This paper deeply characterizes these dependencies by means of a Monte Carlo exploration of the design space. As a result of the empirical experiments, an optimal combination of input parameters is proposed to maximally increase the performance without affecting the lightness of the algorithm.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
