Comparison Study of Deep Visual Tracking on Infrared Imagery in a Maritime Environment

摘要

Over the past years, tracking in the visible domain has seen rapid growth by exploiting Deep Neural Network(DNN) based methods, whereas, tracking in the Thermal Infrared (TIR) domain has seen a small interest. Inthis comparative study, we address tracking in a TIR maritime context for surveillance applications. Towardsthis end, we rst compare the performances of traditional Single Object Trackers (SOTs) and recent DNN-basedSOTs on a TIR maritime data set. Following this, we examine the sequences of the TIR data set causingdiffculties for trackers and identify problematic attributes. Firstly, we use a group constituted of recent stateof-the-art DNN-based trackers and another group constituted of traditional trackers not employing DNN-basedmethods, and measure performance using the following metrics: Intersection over Union (IoU), center error,success rate, and robustness. Furthermore, we rank the trackers by taking into account their scores on IoU androbustness. The presented study shows that recent trackers exploiting DNNs methods for tracking perform onaverage better: over 23.6% on IoU and over 15.5% on robustness than their counterparts not utilizing DNNin their tracking process. Moreover, despite the provided improvement by using DNN-based trackers, a failurecase analysis shows that clutter, occlusion handling, low-resolution and scale change of the target, are visualattributes that still remain challenging, requiring further improvement.