An efficient target tracking algorithm based on an imaging sonar was proposed to solve the problem of underwater multi-target tracking. The echo signal model based on signal intensity was established for each pixel point in the acoustic image according to the imaging features of the sonar in order to extract the individual target from the images. The sequential Monte Carlo probability hypothesis density (SMCPHD) filtering was applied to the target states. The Auction track recognition algorithm was used to associate the filtered target states with the identified tracks, so that the multi-target tracking was realized. The simulation analysis of the algorithm showed that the proposed method was more efficient than the multi-target tracking algorithms based on data correlation, eg. joint probabilistic data association (JPDA) and multiple hypothesis tracking (MHT). A field experiment was conducted to collect the sonar data. The tracking trajectories of all the targets were obtained after the target extraction and tracking.