The invention provides a patrol robot attitude correction method. The patrol robot attitude correction method comprises the specific steps of: performing video camera calibration according to transformation relations between an image coordinate system and a video camera coordinate system and between a camera coordinate system and a world coordinate system, and acquiring video camera internal parameters; carrying out feature matching on images photographed by a patrol robot by adopting an ORB-based feature extraction and matching algorithm, and optimizing the matching result; establishing a pose relation between a video camera and the robot, performing relative pose calculation according to acquired matching point pairs, so as to obtain current rotating angles of the robot, wherein the current rotating angles of the robot comprises an odometer rotating angles and a visual rotating angle; and merging the odometer rotating angles and the visual rotating angle to obtain a current angle tobe corrected of the robot, and completing attitude correction. The patrol robot attitude correction method adopts two optimization methods of distance threshold and cross matching to carry out matching point optimization, and the matching effect is more reliable after optimization.