Detecting multi-interface from the oil-water separation image is the basis of automatic detection of the moisture content and other indicators based on machine vision in heavy oil. Because of the strong viscosity, heavy oil easily attaches to the test tube wall and the oil-air and oil-water interface information is fuzzy. It is difficult to extract different location of the interface information simultaneously in vitro for traditional image interface detection method. To solve these problems, we proposed a difference statistics based method to detect multi-interface in vitro. Based on the interface enhancement of oil-water image, we calculate row average grey value of the oil-water image so that convert the two-dimensional image information into one-dimensional signal, then analyze the difference statistics information of one-dimensional signal, finally locate the position of the oil-air and oil-water interface according to the local extreme value of the difference. Experiments show that the multi-interface detection method has high precision and reaches the requirements of industrial applications, which could be used for a similar scene multi-interface detection.