<p>This paper presents a novel consensus Gaussian mixture probability hypothesis density (GM-PHD) filter to enhance state estimation accuracy for time-varying targets using multiple sensors. The consensus GM-PHD, which relies on the modified Bayesian Consensus Filter (BCF) with the logarithmic opinion pool, is employed to fuse local intensities and achieve the consensus intensity after BCF iterations. Rigorous theoretical proofs are given to guarantee the consistency of the BCF iterations. To alleviate the computational burden, a sampling-based method is further proposed which only updates the particle weights during BCF iterations. Subsequently, particles are resampled based on their weights and then the Expectation-Maximum algorithm reconstructs the Gaussian Mixture Model utilized in the GM-PHD filter from the resampled samples. Simulations involving linear and nonlinear target systems demonstrate the superior performance of the proposed approach in multi-target tracking accuracy compared to existing methods.</p>