In this paper, we consider the distributed Kalman filtering problem over sensor networks with a finite number of sensor nodes where each node can communicate only with its neighboring nodes. We first introduce the concept of gradual information fusion estimation (GIFE) and propose an algorithm for computing the GIFE that can obtain performance improvement via reducing estimation error covariance. Also, we prove that the GIFE can be expressed as a weighted sum of local estimates. Then, based on the results of GIFE and some results proposed in this paper, two distributed Kalman filters are developed where, at time step k, each node is allowed to communicate with its neighboring nodes at most once in the first filter and each node is permitted to communicate with its neighboring nodes twice in the second filter. In addition, we prove that either of the two proposed distributed Kalman filters is unbiased and the estimation error covariance of the first distributed Kalman filter is less than or equal to that without using information fusion estimation. We prove that the estimation error covariance of the second distributed Kalman filter is less than or equal to that of any local estimate belonging to a set. An example of unmanned ground vehicle is provided to illustrate the performance of the two proposed distributed Kalman filters.