To improve the accuracy and computational performance of tunnel crack size measurement， a visual width measurement method based on multi-scale feature fusion of the central axis was proposed. Firstly， the crack width was selected as the research object， and the extraction method of the crack central axis was improved. A template matching strategy was proposed to improve the simplification rate of the central axis pixels generated by the parallel thinning algorithm. Secondly， an optimized central axis branch removal method was proposed to improve the removal effect of redundant branches in the crack skeleton. Finally， the least square method was used to fit the central axis curve， and a width measurement method based on multi-scale direction feature fusion of the central axis of the crack skeleton was proposed. Using the field-of-view angle calibration of a complementary metal-oxide-semiconductor （CMOS） sensor， the pixel width of the crack is converted to the actual crack width. The Huaxi Tunnel in Banan District， Chongqing， was selected as the research object for case analysis. The experimental results showed that， compared with the maximum expansion circle method， the shortest path method， and the orthogonal projection method， the proposed approach reduces the average pixel absolute error of crack width measurement by 9.89 percentage points， 4.67 percentage points and 15.61 percentage points respectively. Moreover， the improved crack axis extraction method achieved a 6.58 percentage points higher pixel simplification rate than that the original method. At the same time， in the crack measurement experiments at different measurement distances， the adaptive ability of the proposed method at different measurement distances was verified.