Cooperative vehicle safety systems rely on vehicular networking for vehicle tracking and collision warning. The most pressing challenge in such systems is to maintain real- time tracking accuracy while avoiding network failure and congestion. In vehicular networking, vehicle density is changing rapidly. This unique characteristic can cause network disconnection or channel congestion. Moreover, the interference factors, such as hidden nodes, can cause the network performance to deviate from the ideal state, which heavily degrades the performance of vehicle tracking. To overcome the above two problems, in this paper, an adaptive power control framework for real-time vehicle tracking under the condition of dynamic vehicle density and interference factor is proposed. The framework consists of two parts: a prescriptive reference model and an adaptive power control model. The prescriptive reference model is used to predict, in a rolling-horizon manner, the desired network state based on the desired tracking accuracy by considering the dynamic vehicle density. The adaptive power control model integrates the desired network state and the current real network state that may be affected by interference to generate real-time power control strategy for accurate vehicle tracking. Experimental results show that the proposed framework can significantly improve the performance of real- time vehicle tracking.