Abstract: Autonomous detection technology is the key technology to realize underwater intelligent unmanned vehicle, and it is the premise that the unmanned vehicle can independently perform underwater early alert and target tracking. Aiming at the problem that the performance of the existing autonomous detection method based on the mean level constant false alarm rate (CFAR) and the ordered statistics CFAR is degraded under the condition of background fluctuation and multi-target, which is due to the inaccurate estimation of background noise statistical characteristics, an autonomous detection method based on the azimuth-time two-dimensional reference window associating the ordered truncation average (OTA) algorithm is proposed. In this method, the azimuth-time two-dimensional reference window is designed to solve the problem of noise inaccurate estimation caused by less reference samples in one-dimensional reference window, and the ordered truncate average algorithm is used to estimate the background noise statistics and to normalize the fluctuating background. Then, a constant false alarm detector is constructed by using the mean value and variance, and a tracking-before-detection technique is adopted to achieve multi-target automatic detection and tracking in the fluctuating background. The lake-test results show that the autonomous detection method has a good effect on multi-target detection under the interference of UUV self-noise.