To provide an accurate operating pose for an air-inlet inspection robot， a vision-based detection and alignment method based on line-feature similarity in the region of interest （ROI） is proposed. Scene images are acquired by a depth camera mounted on the end effector of a mobile manipulator. A technical framework combining deep-learning-based object recognition and machine-vision-based contour measurement is innovatively designed. First， a YOLOv8 network is used to identify the ROI containing the inlet opening from scene images. Second， an improved random line detection （RLD） algorithm is designed to detect the contour lines of the inlet opening within the ROI. Furthermore， a quadrilateral contour extraction method for the inlet opening is proposed， in which the detected lines are refined and spliced based on line-feature similarity， and the quadrilateral contour of the inlet opening is extracted through line-end connection checking， contour-closure screening， and contour-area screening. Based on the extracted contour， a relative pose estimation model of the mobile robot is established， and a pose-feedback controller is designed to accomplish workstation alignment. A simulated experimental system for air-inlet visual alignment is built， and experiments on ROI recognition， line detection， contour extraction， and alignment control verify the feasibility and effectiveness of the proposed method.