The Internet of Vehicles has increased the demand for obstacle avoidance algorithms for autonomous vehicles in dynamic scenarios. Artificial potential field method has become a local obstacle avoidance scheme adopted by many autonomous vehicles thanks to its advantages such as the removal of the need for drawing the map, low computational cost and high real-time performance. However, when facing a larger obstacle perception radius in the environment of IoV, it may mistakenly consider remote targets without collision risk into obstacle avoidance calculation. In addition, when facing a dynamic obstacle, the direction of avoidance may be the same as the movement of the obstacle, resulting in redundant obstacle avoidance or even collision. Therefore, it is necessary to transform the traditional method to adapt to new scenarios. An improved artificial potential field method based on Kalman filter prediction is proposed in this paper, which can avoid the calculation of possible targets without collision risk in a large perception range, and can produce a better path and higher security in the face of dynamic obstacle avoidance. In this paper, extensive simulation experiments under various dynamic scenarios are carried out and demonstrate that our method performs better and avoids collisions where original method fails.