Enhanced Visual Object Tracking and Segmentation in Military Environments: Overcoming Camouflage and Deformation Challenges

Abstract

Visual object tracking is a critical component of border surveillance technology, complementing object detection in its importance. Despite significant advancements enhancing tracking performance, challenges such as object drifting and discrimination among similar objects persist. This is particularly problematic in military settings where distinguishing between soldiers with matching attire is arduous. This paper introduces an innovative model capable of executing visual object tracking and segmentation in tandem. The model's update mechanism allows for sustained tracking, adeptly handling significant variances in the initial bounding box. Enhanced tracking of camouflaged soldiers was achieved through the incorporation of specialized learning datasets focused on camouflage patterns. Testing our model on both standard benchmarks and tailored military datasets yielded impressive results, affirming the model's efficacy.