Scene complexity refers to the difficulty of human perception and understanding of the specific environment. An environment is complex when it has many parts or components, and those parts or components interrelate with each other in multiple and random ways. People’s daily behaviors and spatial activities are often influenced by such complexity of real-world environments. The existing computational methods on modelling the perceived scene complexity of humans primarily focus on either visual or structural characteristics of the environment. This work presents a computational method to quantify the scene complexity of real-world environments comprehensively based on the visual, structural, and semantic characteristics, and assesses the performance of the technical approach with human-labelled “ground-truth” data. Specifically, we proposed a set of features to model the visual (e.g. color, shape, texture, and field of vision), structural (e.g. branches and nodes), and semantic (e.g. the number of POIs and their spatial patterns, well-known signs) aspects, based on street-view panoramas, road network data, POI data, and building footprint data. The results of the evaluation show that the proposed computational method is feasible to predict the perceived scene complexity of street-view environments, with an excellent Mean Absolute Error (MAE) of 0.1108 (on the scale of 1 to 5). The evaluation results on two additional cities further illustrate the high robustness of the proposed computational method. Regarding all conceivable combinations of the visual, structural, and semantic dimensions, considering all these three dimensions provides the best regression performance. In addition, the top-4 most important features for the modelling of scene complexity were: Spatial distribution of POIs, Number of POIs, Percentage of visible sky area, and Distance to the nearest street intersection. Interestingly, these 4 features all appeared in the top-5 feature list reported by human participants in the empirical studies. The employment of openly available data sources makes the proposed method widely applicable to many different cities in the whole world.