Abstract: Manual detection is the main disease detection way for bridge underwater structures, which has the problems of high risk, low efficiency and low accuracy. In response to the above problems, this paper proposes an underwater robot bridge detection method with high-definition underwater camera. The surface detection of concrete structures was carried out in clean water experiment. The image with cracks was analyzed by image processing, Gaussian filtering and gradient calculation of the weight threshold method, and the crack edge features were extracted; The calculation method of crack width under Euclidean distance based on slope calculation was adopted to realize the analysis and calculation of crack width. The experimental results show that the accuracy of crack detection results in this paper is high, and the calculation accuracy of crack width parameters can meet the requirements of code.