This paper studies the distributed filtering of noisy measurements of one scalar quantity. The considered network is composed of two sets of nodes: sensing nodes which perform the measuring task and non-sensing nodes which mediate between the sensing nodes. Inspired by Bayesian sensor fusion, three consensus-based algorithms are proposed. In the algorithms, graph edge weights in effect are determined based on variances of measurements of all the sensors. Evolution dynamics of the expected values and covariances of the state estimates throughout the network is analyzed. Numerical simulations are given to examine the fusion performance of the proposed scheme. The results are fairly consistent with the analytical solution regarding the statistical properties of the steady-state state estimates in the network.