This paper proposes an infrared object tracking algorithm based on two-stage spatiotemporally weighted features. First, the object area is divided into non-overlapping areas of the same size, and different weights are assigned to different location information, from which an adaptive spatiotemporal weighted Bayesian classifier is derived. An improved metric is then used to identify classification samples with the maximum class difference, which have high tracking adaptability, and to enable re-capture and tracking when the target is occluded. Simulation experiments show that, compared with mainstream tracking algorithms such as SiamFC, the proposed algorithm achieves significant improvements in overlap rate and central error indicators on the LSOTB-TIR target tracking dataset, significantly enhancing tracking stability and positioning accuracy. The tracking speed reaches 56 F/s, making it suitable for engineering applications.