In the online detection process of the bearing hole of the trajectory control part of a beverage filling equipment without fixture positioning, an single picture isnt enough for the high-precision detection measuring. This paper proposes a 2 steps method: feature recognition from the overall visual and high resolution local image obtaining by image stitching. Firstly, the feature description matrix is constructed by the central position relationship of the global image bearing hole, and the overall feature hole recognition is performed by the support vector machine (SVM) method. Secondly, for the identified bearing holes, the KAZE method is used to splicing the local feature hole images to achieve high-precision measurement of the bearing holes. Research examples show that the method can quickly achieve high-precision measurement of bearing bores with high measurement efficiency and success rate.