Abstract: As GPS signals are blocked in indoor environments,a vision-based accurate indoor positioning algorithm is proposed referring to fire safety evacuation signs which are widely and evenly distributed in indoor environments.The algorithm aims at calculating distance to the nearest fire safety evacuation sign in the map from the pose of current posi-tion.Color character of fire safety evacuation signs is used for color threshold segmentation.Histogram of Oriented Gra-dient(HOG)features and Support Vector Machine(SVM)are combined to check whether the candidate box contains a fire safety evacuation sign.Holistic Speeded Up Robust Features(SURF)is used for matching,and K-Nearest Neighbor (KNN)method is uses to select nearest K positions as candidate locations.SURF local feature is used for feature matc-hing,a location with the largest number of local feature matches is selected as the result of image-level positioning,and the pose of the current location is calculated in the map.Through the field test in an underground parking lot and an office building,the results show that the proposed method can meet the requirements of accurate indoor positioning,with the accuracy is above 96%,and the average positioning error is below 0.6 m.The results show that this proposed method provides a robust and accurate solution for indoor positioning.