In the unknown target number environment, a new track-before-detect(TBD) algorithm based on cardinalized probability hypothesis density(CPHD) filter is proposed for the tracking and detection of multiple dim targets to avoid the low tracking robustness and high computational amount. The Gaussian particle filter(GPF) approximates posterior densities as Gaussians, the mean and covariance of each Gaussian components in CPHD can be operated recursively by using the particle filter, and particle resampling is not required, which can avoid the problems of particle degeneracy and sample impoverishment. Meanwhile, the updated expression for calculating the particle weight is derived according to the actual TBD situation. Simulation results show that the proposed algorithm is able to convey the cardinalized information more reliably and lower computational time with better tracking performance in number and states of multiple dim targets estimation than the conventional algorithm.