To overcome single-modal tracking limitations in adverse environments, we exploit complementary information from visible (RGB) and thermal infrared (TIR) modalities. However, existing visible and thermal infrared (RGB-T) tracking frameworks often inadequately leverage inter-modal correlations or efficiently mitigate computational overhead from dual-modal fusion. We propose an efficient RGB-T tracker with dual-template cross-modality interaction and foreground selection. The template images from the two modalities are fused to construct a merged template branch, and both the fused template features and the individual modal template features are used as a bridge for cross-modal interaction, thereby addressing the center misalignment between modalities and fully leveraging information from both sources. To reduce computational burden from Vision Transformer(ViT)'s multi-head attention, we construct a polarity-aware linear attention Transformer encoder. Additionally, a foreground selection module processes PolaFormer's attention maps to eliminate background features, enhancing precision while lowering computational load. On the LasHeR dataset, the proposed method achieves success rate and precision of 57.1% and 71.2%, respectively. This represents an improvement of 1.1% and 1.5% over TBSI, with a tracking speed 3.5% higher, surpassing state-of-the-art RGB-T tracking approaches.