Predicting the future trajectory of dynamic objects in a scene is a key capability to ensure safe and smooth driving of autonomous vehicles. Accurately predicting the trajectory of objects is not a simple task due to the complexity of urban scenes and the variety of dynamic object categories. Because objects in a scene are affected by the motion of other objects as well as by static scenes, much of the previous work focused on capturing interactions between objects. In this paper, we combine the map attention mechanism to extract map features, so that the model focuses on the regions in the map that have influence on object motion, thus improving the prediction results. To evaluate its performance, an urban scene trajectory prediction datasets containing map data is produced and we compared with existing prediction algorithms on this datasets.