Target center positioning is a critical technology in the calibration process of infrared thermal images. Given the relatively complex morphology of target images, we propose a center positioning algorithm based on improved template matching with self-constructed convolution kernels. First, the algorithm constructs a normalized template with target image features and performs matching operations on subsampled and preprocessed target images to obtain coarse positioning results. Based on the coarse positioning center, the original image undergoes region of interest (ROI) fine matching, and further correction is achieved through a subpixel subdivision algorithm. Ultimately, the precise target center position is determined. This algorithm effectively detects target images with blurring and indistinct edge features, avoiding interference from blurring, occlusion, complex backgrounds, or indistinct features. It demonstrates good robustness, accurately positions the target center, and operates at high speed. Compared to traditional template matching methods like cross-correlation (CCORR), normalized cross-correlation (NCC), and Hough transform, it offers significant improvements and meets the positioning requirements in the automatic calibration process of infrared thermal imagers.