In the Vanet application scenario, due to the high speed motion of the vehicle, the topology of the network keep changing, and most routing protocols need to maintain their neighbor table in time for routing select. Frequent retransmission of data caused by neighbor selection error will result in high time delay and unreliability. Many Vanet classic protocols cannot be applied to all scenarios. For this reason, this paper proposes a hybrid Vannet routing algorithm based on the highway application scenario NDK (Vanet Neighbor Discovery Method By Kalman Filter). The algorithm uses the GPSR (Greedy Perimeter Stateless Routing) idea of the classic geographic location routing algorithm, with the help of Kalman filter prediction model to predict the neighbor node table, and at the same time, the predicted values are periodically modified by roadside device (RSU, Road Side Unit). The result of NS-3 simulation experiments show that, compared with the classical GPSR algorithm and other algorithms based on time and motion, the algorithm has better packet arrival rate and lower transmission delay.