Non-Line-of-Sight imaging is the imaging of scenes outside the line of sight, which can be divided into active Non-Line-of-Sight imaging and passive Non-Line-of-Sight imaging. Active Non-Line-of-Sight imaging uses active light sources and detectors to image Non-Line-of-Sight scenes. Passive Non-Line-of-Sight imaging relies on the light emitted or reflected from the scene, using objects, corners and windows to image the Non-Line-of-Sight scene. The existing methods of passive Non-Line-of-Sight imaging using windows are only for rectangular windows, and need to know the shape, scale and position of the window. However, in real life, windows have different shapes, and it is difficult to know their specific scale and position. Therefore, a passive Non-Line-of-Sight imaging method with known window shapes is proposed. Firstly, analyze the formation principle of the diffuse reflection image of the Non-Line-of-Sight scene illuminated through the window onto the diffuse reflection surface, and construct the model. Secondly, a series of windows with different scales are selected based on the known window shape to construct the visual matrix, and the L0 gradient sparse constraint is used to solve it. Select the window scale corresponding to the minimum mean square error as the final window scale. Finally, refine the window, and optimize the scene and window jointly to obtain the final reconstructed scene. Experimental results show that for windows with different shapes, the proposed method can reconstruct Non-Line-of-Sight scenes when only the shapes of windows known and realize the Non-Line-of-Sight imaging. Compared with other methods, the average PSNR of the reconstructed imaging results is increased by 0.461dB, and the average SSIM is increased by 0.0361.