Infrared small target tracking on complex backgrounds is a challenging research hotspot, which plays an important role in target warning, ground monitoring and flight guidance. In an infrared background image, the target occupies few pixels and has low contrast and no specific shape; the target edge is not obvious, lacks texture and is point-like; and there are often difficulties in the tracking process such as target occlusion, background clutter and rapid movement, which increase the difficulty of tracking infrared weak and small targets. In order to improve the tracking performance of traditional correlation filters for small targets, we design a target-tracking algorithm based on the structure tensor, which is a symmetric, semi-positive definite matrix that can represent a certain regularity in the neighbourhood of a point in the space, and the structural tensor feature extraction method is used as an epistemic model of the tracking algorithm, which can adequately describe the characteristics of the small targets that exhibit sudden changes in grey scale and are not correlated with the neighbourhood. Compared with origional correlation filters, it has a great advantage in extracting the structural information of the small target. Experiments show that, compared with other tracking algorithms, our proposed tracking algorithm effectively avoids the influence of false alarms in infrared target tracking under complex environments and has a significant advantage in terms of its accuracy index and high computational efficiency.