Periodic Report Summary 1 - VGI_SLAM (Deriving Spatial Data from Volunteered Geographic Information)
Spatial information is a fundamental component of many applications such as navigation aids. However, in order for such applications to be practical the spatial information in question must be of sufficient quality. Spatial information quality lies along multiple dimensions however the relevant dimensions ultimately depend on the nature of the data and intended application. Toward understanding the factors influencing spatial information quality, we consider the two predominant processes by which this information is created.
The first process is through the use of traditional mapping practices which are employed by trained professionals working for proprietary data vendors or national mapping agencies. The second is through crowd-sourcing which is employed in projects such as OpenStreetMap (OSM). Depending on which of these processes is employed, a number of different factors may influence the corresponding spatial information quality. For example, contributors to crowd-sourcing projects generally do not have any formal training with respect to good mapping practices and this can negatively affect logical consistency. Also, individuals working for proprietary data vendors and national mapping agencies generally will not have in-depth geographical knowledge of the areas they map and this can negatively affect attribute accuracy. However, irrespectively of which processes is employed, one factor which influences quality with respect to multiple dimensions is that creating spatial information is labour intensive. If the above mapping processes could be automated to some degree this would reduce labour requirements and in turn positively impact information quality.
Geographical Information Science (GIS) and robotics have traditionally functioned as distinct research domains with a corresponding distinct body of knowledge existing within each. A close examination reveals that both fields share a fundamental research goal. That is, to model the geometry of a space; or put simply, to create maps of an environment. As such there exists great potential for the flow of information between these domains. In the robotics domain there exists a huge body of knowledge regarding how to effectively reasoning in a probabilistic manner about noisy sensor measurements of an environment toward mapping that environment. This is known as the problem of Simultaneous Localization and Mapping (SLAM). The main goal of this project is to apply this knowledge in the development of methodologies for reasoning in a probabilistic manner about crowd-sourced spatial information toward the automation of the corresponding mapping process.
Our efforts so far have focused on the development of a methodology for automating the process of adding semantic information to street networks. In this context we define semantic information to be the class information of individual objects. The proposed methodology is based on the assumption that semantic information is implicitly represented in the corresponding geometrical representation of the street network. Toward justifying this assumption consider Figure 1 which displays the OSM street network for the city of Boston. It is evident that primary streets (represented in red) exhibit characteristics of being quite long, linear and have dual lanes. On the other hand, secondary streets (represented in blue) also exhibit characteristics of being quite long and linear and may or may not have dual lanes. Residential streets (represented in green) exhibit a grid like pattern. Pedestrian streets (represented in grey) exhibit characteristics of being quiet short and non-linear. These observations imply that if one could automatically extract such geometrical patterns or characteristics, it would be possible to use such information to infer the semantic type of individual streets. Toward this goal we extract geometrical patterns from the street network using an advanced probabilistic reasoning paradigm known as probabilistic graphical models. This is a standard approach to formulating problems in the robotics domain. These patterns are then used to determine the semantic type of different streets. The initial results achieved using this approach have been very positive and form the content of a journal paper in the International Journal of Geographical Information Science which is currently under revision.
(see attachment)
Figure 1. A street network is displayed where the corresponding colour of each street segment indicates its semantic type.
The result of this project will be the ability to create spatial data of high quality at lower cost. This will have a number of positive impacts. It will increase the feasibility and scope of research projects using spatial data. The availability of such data will reduce business costs and improve competitiveness. It will also remove the redundancy of capturing the same data multiple times and consequently promote a better use of resources. It has been demonstrated that the EU exhibits Data-divides, defined as the disparity in availability of data for scientific enquiry and decision-making most felt in low-and middle income countries. Through reducing the cost of creating spatial data, the project in question can help reduce such data divides.
The first process is through the use of traditional mapping practices which are employed by trained professionals working for proprietary data vendors or national mapping agencies. The second is through crowd-sourcing which is employed in projects such as OpenStreetMap (OSM). Depending on which of these processes is employed, a number of different factors may influence the corresponding spatial information quality. For example, contributors to crowd-sourcing projects generally do not have any formal training with respect to good mapping practices and this can negatively affect logical consistency. Also, individuals working for proprietary data vendors and national mapping agencies generally will not have in-depth geographical knowledge of the areas they map and this can negatively affect attribute accuracy. However, irrespectively of which processes is employed, one factor which influences quality with respect to multiple dimensions is that creating spatial information is labour intensive. If the above mapping processes could be automated to some degree this would reduce labour requirements and in turn positively impact information quality.
Geographical Information Science (GIS) and robotics have traditionally functioned as distinct research domains with a corresponding distinct body of knowledge existing within each. A close examination reveals that both fields share a fundamental research goal. That is, to model the geometry of a space; or put simply, to create maps of an environment. As such there exists great potential for the flow of information between these domains. In the robotics domain there exists a huge body of knowledge regarding how to effectively reasoning in a probabilistic manner about noisy sensor measurements of an environment toward mapping that environment. This is known as the problem of Simultaneous Localization and Mapping (SLAM). The main goal of this project is to apply this knowledge in the development of methodologies for reasoning in a probabilistic manner about crowd-sourced spatial information toward the automation of the corresponding mapping process.
Our efforts so far have focused on the development of a methodology for automating the process of adding semantic information to street networks. In this context we define semantic information to be the class information of individual objects. The proposed methodology is based on the assumption that semantic information is implicitly represented in the corresponding geometrical representation of the street network. Toward justifying this assumption consider Figure 1 which displays the OSM street network for the city of Boston. It is evident that primary streets (represented in red) exhibit characteristics of being quite long, linear and have dual lanes. On the other hand, secondary streets (represented in blue) also exhibit characteristics of being quite long and linear and may or may not have dual lanes. Residential streets (represented in green) exhibit a grid like pattern. Pedestrian streets (represented in grey) exhibit characteristics of being quiet short and non-linear. These observations imply that if one could automatically extract such geometrical patterns or characteristics, it would be possible to use such information to infer the semantic type of individual streets. Toward this goal we extract geometrical patterns from the street network using an advanced probabilistic reasoning paradigm known as probabilistic graphical models. This is a standard approach to formulating problems in the robotics domain. These patterns are then used to determine the semantic type of different streets. The initial results achieved using this approach have been very positive and form the content of a journal paper in the International Journal of Geographical Information Science which is currently under revision.
(see attachment)
Figure 1. A street network is displayed where the corresponding colour of each street segment indicates its semantic type.
The result of this project will be the ability to create spatial data of high quality at lower cost. This will have a number of positive impacts. It will increase the feasibility and scope of research projects using spatial data. The availability of such data will reduce business costs and improve competitiveness. It will also remove the redundancy of capturing the same data multiple times and consequently promote a better use of resources. It has been demonstrated that the EU exhibits Data-divides, defined as the disparity in availability of data for scientific enquiry and decision-making most felt in low-and middle income countries. Through reducing the cost of creating spatial data, the project in question can help reduce such data divides.