Sketch maps are one of the most commonly used instruments to study people's perception of the environment. While previous research has focused on analyzing sketch maps with respect to their accuracy, completeness, or distortion, there is little study focusing on generalization (the level of geometric details of the depicted features) in sketch maps. Furthermore, when investigating people's mental representations of space, researchers compute the accuracy or completeness of sketch maps via a one-to-one comparison of sketched features with features in metric map data. The task of one-to-one comparison becomes problematic because features in sketch map often represent many features in a metric map. This paper identifies the frequently occurring generalizations in 108 sketch maps of small urban area by manually extracting generalized features such as streets and buildings and classifying them based on similarities. We found three generalization types in streets, two in junctions and four in buildings. To evaluate our classification, we created 10 sketch maps by systematically introducing the nine identified generalizations. A study was conducted wherein five raters were briefed about the different generalization types. Each of them classified the features in sketch maps as generalized and non-generalized. We found a high inter-rater agreement, implying the comprehensibility and clarity of the identified generalization types. We propose that such systematic view on the generalization in sketch maps would standardize the process of comparing them with metric map data as well as enable a new way of analyzing sketch maps with respect to generalization.

草图是研究人们对环境的感知最常用的工具之一。虽然以前的研究集中在分析草图的准确性、完整性或失真，但很少有研究关注草图中的概括（所描绘特征的几何细节水平）。此外，在调查人们对空间的心理表征时，研究人员通过将草图特征与度量地图数据中的特征进行一对一比较来计算草图的准确性或完整性。一对一比较的任务变得有问题，因为草图中的特征通常代表度量图中的许多特征。本文通过人工提取街道、建筑物等泛化特征并根据相似度对其进行分类，识别出108幅小城区草图中的频繁泛化。我们在街道中发现了三种泛化类型，在路口发现了两种，在建筑物中发现了四种。为了评估我们的分类，我们通过系统地介绍九个已确定的概括创建了 10 个示意图。进行了一项研究，其中向五位评估者简要介绍了不同的概括类型。他们每个人都将草图中的特征分为广义和非广义。我们发现了一个很高的评分者间一致性，这意味着所识别的泛化类型的可理解性和清晰度。