As an important tool for underwater resource development, the relative positioning of underwater vehicles plays an important role in performing complex tasks. The underwater manta ray like robotic fish ( UMLRF) obtains forward power by swinging its pectoral fins, and it has the characteristics of flexible movement and few fixed points of the carrier structure. Therefore, it is difficult to estimate the relative position by collecting fixed points with visual sensors. Because of the above problems, we use the deep convolutional neural network method to directly detect the target of the UMLRF without installing fixed beacons. Then, the binocular camera is used to measure the relative distance between the UMLRF, and finally, the target detection and distance measurement results are given through the underwater experiment.