PROJECT TITLE :

RGBT Tracking via Noise-Robust Cross-Modal Ranking

ABSTRACT:

The currently available RGBT tracking methods usually involve the use of a bounding box to localize a target object. In this method, the trackers are frequently impacted by the presence of background clutter. This article presents a novel algorithm to suppress background effects in target bounding boxes for RGBT tracking. The algorithm is called noise-robust cross-modal ranking, and it was developed in order to address this issue. In particular, we deal with the noise interference in cross-modal fusion and seed labels by addressing it from two different vantage points. First, the concept of "soft cross-modality consistency" is put forward as a way to allow for "sparse inconsistency" when fusing different modalities. This is done with the intention of taking into account both the collaborative nature and the heterogeneity of the various modalities in order to achieve more successful fusion. Second, the optimal seed learning is designed to handle the label noises of ranking seeds that are caused by certain issues, such as irregular object shape and occlusion. This is accomplished through the use of a neural network. In addition, in order to implement the complementarity and keep the structural information of the various features within each modality, we perform an individual ranking for each feature and employ a cross-feature consistency in order to pursue their collaboration. This helps us both deploy the complementarity and maintain the information about the structures of the different features. In order to solve the proposed model, a unified optimization framework that has an effective efficient convergence speed has been developed. Extensive experiments show that the proposed method is effective and efficient when compared to other tracking methods that are considered to be state-of-the-art when using the GTOT and RGBT234 benchmark data sets.