Abstract: This paper proposes an optimized matching idea for a speeded up robust features (SURF) algorithm with a rotation invariance that can solve the problems associated with low matching accuracy, which are caused by rotation and pitch angle deviation in the multi-angle recognition of an object.In this paper, we utilize the clustering method to classify the extracted key points.The key points are then gained on the horizontal and vertical Haar wavelet values by distance Gaussian weights to more accurately determine the main direction of the feature points.Simultaneously, aiming at the mismatching pairs appearing in the matching process, we remove them by introducing the wrong match rough reduction method.In addition, to further improve the matching opportunities, we view the panoramic image of the object as the background reference image.Experimental results show that the algorithm has a greater probability and recognition accuracy with regard to multi-angle image recognition of objects and consumes less time.It also has a certain degree of practicality and generalization.