德国慕尼黑大学空间智能实验室

Spatial Intelligence Lab

（http://www.uni-muenster.de/Geoinformatics/en/sil/research/）

 

实验室研究概况

       空间智能实验室围绕空间信息智能表达处理，进行多学科交叉研究。重点研究人们在使用结构化空间知识过程中的认知方法，并利用这些认知规律设计更好的地理信息系统交互。当前主要研究领域包括：

• 识别、校准草绘地图（Recognition and alignment of sketch maps）;
• 寻路辅助的认知和空间学习（Cognitive wayfinding assistance and spatial learning）;
• 表达空间的模糊性（Representing spatial vagueness）;
• 移动设备地理信息可用性（Usability of mobile and tangible devices for geospatial information）

 

研究方向一：识别、校准草绘地图

 

       空间智能实验室的识别、校准草绘地图研究项目属于DFG资助的Sketch Mapia 项目（www.sketchmapia.de）的一部分。SketchMapia项目的目的是开发一套草绘地图采集、识别、解释、集成和可视化框架。根据VGI的思想，SketchMapia利用草绘地图来贡献地理信息。该项目项公众开发了许多地理信息系统功能。尽管人们在绘制地图时存在很多困难，一些用户仍可以利用SketchMapia的接口生产用户所需的空间内容。然而，由于人们认知的差异，草绘地图通常不完整、变形、图形化，因此草绘地图没有矢量地图那么精确。SketchMapia项目提出了一种定性计算模型以计算机可理解的方式表达草绘地图。SketchMapia将多种草绘地图和矢量图集成到一个数据仓库中，用户可以通过查询接口进行访问。在该项目涉及到草绘对象的语义识别、草绘地图校准的认知规则、定性的形式描述、草绘地图的计算模型。

       针对草绘地图对象语义识别，空间智能实验室设计了从影像中提取对象位置和语义的算法，涉及低层、中层影像处理技术和高层影像理解方法。草绘地图可以看作是认知地图的客观表现。因此，草绘地图跟认知地图具有相似的特征：图形化、方向扭曲、距离、大小、形状。本研究中分析了根据矢量地图表达的实际空间进行草绘地图自动校准的可靠性和精度。可靠的定性草绘表现在方向、距离、拓扑、序列、循环顺序等可以利用现有定性计算方法提取出来并形式化描述。SketchMapia目的在于定性表达，增强图式化、扭曲，提供了从矢量地图进行草绘地图空间对象的定性校准基础方法。一组草绘地图的定性校准涉及定性约束网络的匹配，满足最大可能的定性约束。跟矢量地图相比，草绘地图通常会绘制多个抽象层地图，矢量地图中的聚集区域往往是独立的，矢量地图中的聚集空间关系也更精确。因此，需要容错匹配方法来处理不同层次的抽象，以提升校准的质量。

 

研究方向二：寻路辅助的认知和空间学习（Cognitive Wayfinding Assistance and Spatial Learning）

 

       通过寻路认知研究，设计一些新的路径规划算法，方便的计算出包含路径描述指示信息的路径。然而，当前的研究仍然遵循传统导航系统的规则：给出路径和向导信息，用户需要一步一步去寻找。相反，构建一系列逻辑向导信息嵌入在整个任务重，每个向导信息都是独立的，让信息内容最小化。用户寻路任务会减少，只需在指定的位置执行预定义的操作就行。

       新的研究方向包括设计一套新的寻路辅助系统，能够支持获取空间知识和认知地图绘制，以提升用户在不熟悉环境中的方向感。寻路向导信息将被嵌入在环境上下文和整个任务中。丰富的向导信息能够与用户的认知地图进行关联，帮助用户获取和保持方向感。这能够使得用户寻路更容易成功，在运动过程中获取空间决策、走捷径或者自发的绕行。这项研究明确了哪些上下文能够增加用户的方向，这些信息怎样进行表达。

 

研究方向三：表达空间模糊性（Representing Spatial Vagueness）

 

       理解和表达没有明确地理边界的区域成为地理学家关注的研究内容。尽管人们每天用直觉描述不确定区域，但用语言、数字方式表达这些位置仍存在很多困难。者对于地理应用、AI或予以Web中进行空间推理任务非常重要。空间智能实验室的研究分析从用户需求的角度怎样对现有方法进行分类。目前感兴趣的方法包括概率、模糊方法、粗糙集、蛋黄（egg-york）模型、赋值语义、三角网等，来描绘这样的不确定区域。

       此外，该实验室还调研了通过表达用户具体的属性和使用目的能改进模型中哪些区域的语义。要解决的问题是怎样表达这些模糊区域的上下文。

 

研究方向四：移动触控设备地理信息的可用性

 

       这项研究的目前是获取和表达空间信息的语义，以便进行高效、精确的地理信息处理。需要采用一些智能的方法满足用户需求并解决语义一致性问题。这项研究也关注空间数据语义注释，并设计一种模型来度量自然语言表达的语义相似度。

       在人和机器之间进行完美通信的前提是计算机能够正确处理指令。同时机器利用定义好的语言和形式化规则来处理信息，人更喜欢包含模糊语义的自然语言表达。这项研究将通过实验调研自然语言空间关系的语义，设计一种支持语义和空间关系推理的计算模型。自然语言关系和符合认知规律的交互将能改进地理信息系统查询语言，提升可用性。

 

Spatial Intelligence Lab – Institute for Geoinformatics

Who We Are and What We Do

The Spatial Intelligence Lab is part of the Institute for Geoinformatics at the University of Muenster and deals with research problems from the interdisciplinary field of geographic information science, computer science and cognitive science.

Our research focuses on the investigation of intelligent representation and processing of spatial information as well as the examination of human strategies to acquire and organize knowledge about spatial environments. We investigate experimentally human abilities of spatial cognition and cognition of complex spatial environments. At present, particular focus lies on the recognition of real-world objects represented by sketch drawings and the investigation of cognitive maps via analyzing sketch maps. Based on these psychological findings, we develop intelligent computational systems and novel strategies for knowledge processing that are more similar to the human way of thinking. We aim at making spatial information processing more efficient and the interaction with GI software easier. Usability studies are applied to evaluate our new approaches for their usability.

 

Research

Our research brings together topics from a diverse array of subject areas under the unified theme "intelligent representation and processing of geospatial information" . We are particularly interested in understanding the techniques that human agents employ to structure spatial knowledge and use this understanding to provide better methods for interacting with GI systems. Our current research topics are grouped into four main areas:

• Recognition and alignment of sketch maps
• Cognitive wayfinding assistance and spatial learning
• Representing spatial vagueness
• Usability of mobile and tangible devices for geospatial information

Recognition and alignment of sketch maps

SIL's research on recognition and alignment of sketch maps is carried within the DFG funded Sketch Mapia project (www.sketchmapia.de). SketchMapia is a project that aims to develop a framework for collection, recognition, interpretation, integration and visualization of sketch maps. In the context of Volunteered Geographic Information (VGI), SketchMapia employs sketch maps for contribution of geographic information. It opens more capabilities of Geographic Information Systems (GISs) to the general public. In spite people who have difficulties to draw a map, all others can use the sketching interface that SketchMapia provides to produce user-generated spatial contents. However, due to human cognition, sketch maps are incomplete, distorted, schematized, and therefore not as accurate as metric maps. The Sketch Mapia project develops a qualitative computational model to represent sketch maps in a computer-understandable way. SketchMapia integrates information from various sketch maps and metric maps into one data repository which can be queried by users via a query-by-sketch interface. The main research topics within the Sketch Mapia cover the semantic recognition of sketched objects, cognitive criteria for sketch map alignment and their qualitative formalization as well as computational aspects for sketch map alignment.

For the semantic recognition of sketch map objects, we develop algorithms to extract the location and the meaning of objects within the images. On this level the research covers low level and mid level image processing techniques as well as high level images understanding methods. Sketch maps are often considered as externalizations of cognitive maps. Thus, sketch maps appear to have similar characteristics as cognitive maps: They are schematized and distort directions, distances, size, and shapes. . This research investigates the reliability and accuracy of sketching aspects for aligning sketch maps automatically with the corresponding real-world configurations represented on metric maps. Reliable qualitative sketching aspects indicating orientation, distance, topology, serial or cyclic order are extracted and formalized using existing qualitative calculi. SketchMapia aims at qualitative representations that are robust against schematizations and distortions, and provide a basis for qualitative alignment of spatial objects from sketch maps with those from metric maps. Qualitative alignment of a pair  of sketch maps involves matching the qualitative constrain networks of the pair such that the greatest possible number of  qualitative constraints is satisfied. Sketch maps are usually drawn at a more abstract level than metric maps, often aggregating regions that in metric maps would be seperate and aggregating spatial relations that in metric maps would be more precisely determined. Error-tolerant matching methods must therefore be used to account for differences in levels of abstraction inorder to improve the quality of alignment.

Cognitive Wayfinding Assistance and Spatial Learning

Research on cognitively enabled wayfinding has led to new path planning algorithms computing easyto-follow routes with descriptive instructions. Nevertheless, current research still adheres to the principles of traditional navigation systems: routes are given as sequences of instructions that users need to execute step-by-step. Instead of forming a logical sequence of instructions embedded in the overall task, each instruction is isolated and reduced to a minimum of information content. The user’s wayfinding task is cut down to executing predetermined actions at given locations.

This research direction suggests new wayfinding assistance systems that support the acquisition of spatial knowledge and cognitive map-making for advancing the user’s orientation in unfamiliar environments. Wayfinding instructions are to be embedded in the context of the environment and the overall task. Instructions enriched with information that can be related to the user’s cognitive map helps users to get and remain orientated. This makes wayfinding more successful because it enables users to take informed spatial decisions for circumnavigating traffic, taking shortcuts or including spontaneous
detours. Our research determines which context information advances orientation of users and how this information has to be represented.

Representing Spatial Vagueness

Understanding and representing regions which have no well defined boundaries has always been of interest to geographers. Despite being intuitive to humans who use terms to describe vague regions in everyday language, digitally representing such places is a challenge. This is important in order to perform reasoning tasks in geospatial applications, AI or the semantic web. Our research in this direction investigates how existing methods can be classified from the perspective of user requirements. Methods of interest to us include probabilistic and fuzzy methods, rough sets and egg-yolk models, supervaluation semantics, and triangulated irregular networks for delineation of such regions.

In addition to this we investigate where the semantics of such regions can improve the models by presenting users with a view that is tailored to their profile and intended use. We address the problem of how such a contextual view of the vague regions can be developed and  presented to users.

Usability of Mobile and Tangible Devices for Geospatial Information

The research focus on capturing and representing the semantics of spatial information in order to enable effective and accurate information processing. Intelligent methods are required to provide optimal support for users' needs and overcome semantic interoperability problems. Our research is also directed at semantic annotations of spatial data and the development of models to measure semantic similarity of natural language expressions.

Consistent and flawless communication between humans and machines is the precondition for a computer to process instructions correctly. While machines use well-defined languages and formal rules to process information, humans prefer natural-language expressions with vague semantics. We investigate experimentally the meaning of natural-language spatial relations and develop a computational model to specify the semantics and reason on spatial relations. Natural-language relations and cognitively plausible operations shall improve query languages of geographic information systems and increase the usability for humans.