Intention recognition for multiple agents is an important problem in multi-agent systems, and is widely used in autonomous driving, human-machine interaction and military field. Due to the uncertainty in the scale of multiple agents and the distribution of obstacles, the generalization ability of a current intention recognition model is limited. To reduce the sensitivity of the model to the number of agents, an intention recognition algorithm based on a situation map sequence is proposed. The observed information of multiple agents is converted into the situation map sequence and the model is trained based on the situation map sequence. In order to improve the adaptability of the model, the generating method of the obstructive situation map is proposed for situations with obstacles. In addition, in order to reduce the dependence on expert knowledge, the repulsive field factor is estimated using convolutional neural networks. Finally, the proposed method is compared with other methods and ablation experiments are conducted. The accuracy and generalization ability of the proposed algorithm are verified by the results.