A building safety assessment algorithm for self-organizing crowds was proposed, in order to resolve the problems of unreasonable indoor building nodes and the difficulty in choosing escape routes for crowds. Based on IndoorGML, the horizontal plane structure and vertical hierarchical structure of indoor space were established to form a node-relation graph in the topological space, so as to represent the spatial structure of multi-story buildings. A safe evacuation route selection algorithm based on the A^* algorithm was designed, by defining the risk and the traffic rate of unit space. Unreasonable nodes of indoor buildings were found and optimization solutions were proposed, by analyzing the number of people choosing escape routes, the highest risk of nodes on a single escape route, and the average escape time of a single escape route. A simulation scene was constructed for simulation experiments, and the results of different route selection methods were compared to analyze and prove the effectiveness of the proposed assessment algorithm. Results show that the proposed method can be used to find the main factors of building congestion and the degree of influence on evacuation, which is helpful to carry out scientific building safety assessment in the building design stage.