Siamese network-based algorithms have progressively supplanted traditional methods in single-object tracking, offering superior accuracy and real-time performance. Nonetheless, in these tracking algorithms, the temporal context between consecutive frames remains underutilized. This leads to a drift in bounding boxes when similar objects emerge and obscure the original target, thereby adversely affecting tracking performance. To tackle this challenge, we propose a single-object tracking algorithm that amalgamates temporal context and bounding box refinement within the Siamese network framework. Building upon the TransT algorithm, our approach integrates time-adaptive convolution during the feature extraction stage of target tracking to enhance the utilization of temporal context. We introduce Alpha-Refine as a bounding box refinement module to acquire more precise target contour information. The efficacy of our algorithm compared to TransT is substantiated through extensive experiments. Results on challenging datasets, including GOT-10k, TrackingNet, and OTB100, notably excel in tests on the large-scale datasets GOT-10k and TrackingNet benchmarks. Our tracker runs at approximatively 36 fps on GPU.