Infrared object tracking method based on kernel correlation filters

Abstract

Benefiting from the superiority visual capacity of infrared imaging system, infrared object tracking is widely applied in military, weapon guidance, security protection and other fields. Recently, correlation filter has attracted considerable attention within the tracking community due to its high computational efficiency. However, traditional correlation filter based trackers use the single feature, which has poor stability in low resolution and complex background. Aiming at the low resolution, poor contrast, low signal-to-noise and insufficient texture information of infrared image, we propose an Infrared object tracking method based on Kernel Correlation Filters, named Inf-KCF. We design a Multi-Feature Fusion structure to improve the expression ability for infrared targets, and separate the target from the background well. Experimental results on public infrared dataset show that, the Inf-KCF achieves promising performance.