A Fusion based particle filter track-before-detect algorithm (FPF-TBD) is proposed for dealing with dim targets, in which the importance density function of the Particle filter (PF) is generated by means of a fusion algorithm. In order to construct an accurate approximation to the true proposal distribution, the state at each time scan is predicted according to the Extended Kalman filter algorithm (EKF) and the Unscented Kalman filter (UKF) simultaneously. The information, based on the recursion of the weights, is gathered over multiple scans, and the detection decision is made based on tracking results at the end of the processing chain. By making best use of the recent measurements, this new proposed method can obtain an accurate approximation to the system and as a result, improve the track accuracy and detection performance. Simulation results illustrate the effectiveness of this approach.