As a common kind of cooperative target, checkerboard is widely used in camera calibration and position–pose measurement in which the corner detection plays an essential role. However, due to the influence of the complex external environment, the taken checkerboard image is often incomplete, and the pixel coordinates of the detected corners do not correspond one-to-one to their real-world coordinates. To overcome this problem, a robust corner detection method based on backward position matching for incomplete checkerboard is proposed. By employing a decision of the checkerboard edge after the initial corner detection, the prior corner defined by us is first identified, which provides the prior knowledge for the subsequent uncertain corner detection. Then the homography matrix is estimated using the pixel coordinates and the corresponding real-world coordinates of the prior corners. Finally, according to the homography backward mapping, the uncertain corners are successfully determined, and the one-to-one correspondence between the pixel coordinates and the real-world coordinates is further achieved. Experimental results demonstrate that our proposed method is able to identify accurately each detected corner of the incomplete checkerboard and achieves superior robustness in the corner detection under different interference conditions.