Evacuation behavior is an important factor which must be considered in the design of public structures. With the continuous complexity of structure, more and more factors should be considered in evacuation. The traditional design based on experience may have some limitations in practice. Based on the deep neural network model, the evacuation design simulation for subway station buildings is implemented. VR video tracking technologies such as auxiliary image data pre-training algorithm, tracking sequence pre-training algorithm, and recursive neural network model based on You Only Look Once (YOLO) are introduced. Compared with the convolutional neural network (CNN) model, the classified data set pre-training model, and YOLO algorithm, the accuracy and training speed of the model algorithm are verified. In simulation, the Zhujiang New Town Station in Guangzhou is taken as the object. The initial evacuation test point is selected according to the structure of the subway platform, and the test personnel are selected according to the test requirements. The average evacuation time and the average satisfaction score of the testers under the influence factors such as gender, age, subway frequency, and familiarity with VR equipment, as well as under the initial starting points of different evacuation tests. The results show that the accuracy of the algorithm is lower than that of the CNN, but the training speed is faster. The accuracy of the model based on YOLO recurrent neural network is the highest. Although the training speed is 19 ms, which is higher than other models, the overall performance is the best. Differences in factors such as gender, age, frequency of subway ride, and familiarity with VR devices will result in different differences in average evacuation time and average satisfaction score. When the platform center is used as the initial evacuation point, the average evacuation time is the shortest, and the average satisfaction score of the testers is the highest. In conclusion, through VR video tracking technology, the actual situation of subway station buildings can be well simulated, and further design schemes can be made according to the simulated situation, which has practical reference significance.

疏散行为是公共结构设计中必须考虑的重要因素。随着结构的不断复杂，在疏散中应考虑越来越多的因素。基于经验的传统设计在实践中可能会有一些限制。基于深度神经网络模型，对地铁车站建筑物的疏散设计进行了仿真。介绍了VR视频跟踪技术，如辅助图像数据预训练算法，跟踪序列预训练算法以及基于“仅看一次”（YOLO）的递归神经网络模型。通过与卷积神经网络模型，分类数据集预训练模型以及YOLO算法的比较，验证了该模型算法的准确性和训练速度。在模拟中 以广州的珠江新城站为对象。根据地铁站台的结构选择初始疏散测试点，并根据测试要求选择测试人员。在性别，年龄，地铁频率，对VR设备的熟悉程度等影响因素以及不同疏散测试的初始起点下，测试人员的平均疏散时间和平均满意度得分。结果表明，该算法的精度低于CNN，但训练速度较快。基于YOLO递归神经网络的模型的准确性最高。尽管训练速度为19毫秒，比其他型号要高，但总体表现是最好的。性别，年龄，乘坐地铁的频率以及对VR设备的熟悉程度会导致平均疏散时间和平均满意度得分出现不同的差异。当平台中心用作初始疏散点时，平均疏散时间最短，测试人员的平均满意度得分最高。综上所述，通过VR视频跟踪技术，可以很好地模拟地铁车站建筑物的实际情况，并根据模拟情况制定进一步的设计方案，具有实际的参考意义。测试人员的平均满意度得分最高。综上所述，通过VR视频跟踪技术，可以很好地模拟地铁车站建筑物的实际情况，并根据模拟情况制定进一步的设计方案，具有实际的参考意义。测试人员的平均满意度得分最高。综上所述，通过VR视频跟踪技术，可以很好地模拟地铁车站建筑物的实际情况，并根据模拟情况制定进一步的设计方案，具有实际的参考意义。