Aimed at the problem of underwater multi-target tracking using forward looking sonar images, a modified particle filter (PF) multi-target tracking method was proposed. By introducing the joint probabilistic data association (JPDA) method, a JPDA-PF algorithm was established. Thus particle weight could reflect the relationship between measurement and trajectory, which ensured the reliability and continuity of every trajectory. In order to increase the accuracy of tracking, the confirmation matrix was obtained by a double tracking threshold, including both relative distance and angle threshold. The methods of trajectory initialization and trajectory termination were complemented, which improved the whole tracking algorithm. Contrast tests on multi-target tracking were conducted, and the better performance of JPDAPF was verified. This paper provides a more effective method for multi-target tracking based on forward looking sonar.