According to the general problem of unknown detection probability in the probability hypothesis density(PHD) filter, a PHD algorithm based on the time-varying Kalman filter(TVKF) is proposed. Firstly, PHD recursions without the knowledge of the detection probability are derived. Secondly, the measurements of loss events are modeled as Markov processes, and the optimal estimator with missing sensor data samples is given by using time-varying Kalman filter. Furthermore, the closed form solutions are calculated under the framework of the Gaussian sum based probability hypothesis(GMPHD) filter. The simulation results show that the improved algorithm has better performance in terms of state estimation under the unknown detection probability, and has good application prospects.