To address the limitations of target tracking algorithms in long-term tracking scenarios, such as poor adaptability to target scale variations, small target sizes, occlusion, and interference from similar objects, we propose a novel dual-template tracking algorithm with dynamic receptive fields, named static-dynamic template tracker (SDT-Tracker). First, we redesign ResNet50 with a parallel attention mechanism to build a feature extraction network with dynamic receptive fields, enabling efficient feature extraction for the target of interest. Then, we introduce three downsampling methods to capture multi-angle features by integrating local, raw, and key features, thereby minimizing feature information loss. Finally, we propose a static-dynamic dual-template tracking strategy, where the dynamic branch continuously incorporates subsequent frame information, while the static branch extracts the target’s initial information, suppressing irrelevant information introduced by the dynamic branch at key moments. This reduces interference from similar objects and occlusions. Experiments on the LaSOT and OTB100 datasets demonstrate the effectiveness and superiority of the proposed algorithm. Additionally, we deploy the algorithm on the Jetson Xavier NX embedded device for performance testing, achieving a processing speed of 24 frames per second. Compared to classical tracking algorithms, the proposed method shows higher accuracy in complex scenarios and effectively addresses issues of occlusion and interference from similar objects.