In Internet of Vehicles, the difference in vehicle distribution density may lead to uneven load of Mobile Edge Computing (MEC) servers, resulting in hot area servers being overloaded while others being idle. An effective way to solve this problem is task computation through vehicle-MEC and MEC-MEC collaborations. To this end, a Two-stage Stackelberg Game based Task Offloading scheme is proposed, referred as TSGTO. Specifically, we utilize the collaboration between vehicles and MECs for task processing, and allows overloaded MECs to further forward tasks to idle MECs, thereby maximizing resource utilization. First, considering the competition and cooperation between vehicles and servers, we transform task offloading into a two-stage Stackelberg game problem. In the first stage, the vehicle acts as the leader to make a task offloading strategy, determining the optimal offloading ratio based on the internal module dependencies. In the second stage, the MEC server further develops a task forwarding strategy based on the vehicle’s offloading strategy and its own load. Finally, we design a distributed iterative algorithm to obtain the Nash equilibrium of the game. Extensive experiments conducted on real-world datasets demonstrate that, compared with benchmark methods, the TSGTO reduces delay by 5.78%-13.21%, reduces energy consumption up to 15.32%.