This paper is concerned with the state estimation problem for networked systems in presence of packet dropouts in both sensor-to-filter and controller-to-actuator channels. The considered stochastic packet dropouts are independent and Bernoulli distributed. Based on an auxiliary linear-minimum-variance (LMV) filter and hard-decision, a switched filter is proposed for state estimation. The proposed filter is proved to have better estimation performance than the existing LMV filter. A sufficient condition on stability of the switched filter for time-invariant systems with constant controller outputs is also given. Unlike the exiting stability conditions for LMV filters, the proposed one does not require that the estimated plant is stable. Simulation results show the advantage and the effectiveness of the developed filter and its stability condition.