With the development of modern science and economy, congestions and accidents are brought by increasing traffics. And to improve efficiency, traffic signal based control is usually used as an effective model to alleviate congestions and to reduce accidents. However, the fixed mode of existing phase and cycle time restrains the ability to satisfy ever complex environments, which lead to a low level of efficiency. To further improve traffic efficiency, this paper proposes a crowd-based control model to adapt complex traffic environments. In this model, subjects are deemed as digital selves who can perform actions in complex traffic environments, such as vehicles and traffic lights. These digital selves have their own control processing mechanisms, properties, and behaviors. And each digital self is continuously optimizing its behaviors according to its learning ability, road conditions, and information interactions from connections with the others. Without a fixed structure, the connections are diverse and random to form a more complex traffic environment, which may be connected or disappeared at any time with continues movements. Finally, feasibility and effectiveness of the crowd-based traffic control model is demonstrated by comparison with fixed traffic signal control model, indicating that the model can alleviate traffic congestion effectively.