Abstract: Surface defects of underwater structures (e.g.dams, offshore platforms, ports and bridges) pose a threat to their operation and are the main risk factors inducing or causing major accidents. In this paper a method is proposed for measuring the surface defects of underwater structures based on binocular vision, and it is applied to surface crack detection of reservoir dams under deep water. In this method, the parameter matrix of a binocular camera in the underwater environment is first calibrated for correction of camera images, and then a biomimetic technology is adopted for image enhancement. Thereafter, a stereo measurement based on the semi-global matching approach is conducted to obtain a dense disparity map of the defects to deduce their geometric parameters and finally achieve the surface detect measurement of underwater structures. As shown in the experiment, this method can highlight image defects and make measurement results less dependent on the underwater environment; moreover, this method can provide a dense disparity map without invoking scale calibration and is simple in use with a measurement accuracy meeting the engineering requirement. This method enables practical detection for the prevention of disasters and assessment of the underwater structural health of China's major water-resource allocation projects.