Siam-CLF: Object Tracking using Siamese Network
Ashish Kumar Gupta, Deepak Kumar Mishra · 2025
In computer vision, object tracking in thermal imagery has a significant challenge. Object tracking involves finding an object’s location in all video frames based on the first frame location. Thermal object tracking is very useful in bad weather conditions like fog, snow, illumination. The tracking algorithm is more efficient by combining the complementing features from visible and infrared images. The Siam-CLF tracker is an STSO tracker with thermal and RGBT operation modes. The dataset includes thermal and equivalent RGB images. Here, the tracker first generates a fused image from the equivalent feature in the thermal and RGB image and then tracks the object in sequences of fused images. The tracker follows thermal images in the absence of equivalent RGB images in the dataset. A contrastive learning-based network is used for image fusion; the encoder extracts the feature from thermal and RGB images and concatenates the feature. A multi-scale structural similarity-based encoder reconstructs the image from concatenated features. A SiamMask-based network is implemented for the identification of the location of the target object.