For the subpixel edge detection algorithms based on traditional Zernike moment, they often suffer from the difficulties in long time-consuming for manual threshold setting or poor robustness. To address such problems, an improved Zernike moment subpixel edge detection algorithm based on adaptive threshold is proposed in this paper. First, the convolution of a mask of 7 × 7 and the image is calculated to obtain the Zernike orthogonal moments. And then, a three-step grayscale edge model including a transition edge region is established and the calculation formula of the step grayscale is developed. Subsequently, the edge parameters of the model are deduced. Finally, an adaptive threshold is determined by using two-dimensional Otsu method for sub-pixel edge detection. The sub-pixel edge is determined and its sub-pixel coordinates are obtained. The edge detection experiments are performed both on the synthetic image and the real image by using the proposed algorithm and the Zernike moment algorithm with \(7 \times 7\) template respectively. And the experiments results demonstrate that the proposed algorithm has higher location accuracy on the synthetic image. Additionally, the proposed method can have an adaptive threshold and is effective with more robustness and denoising.