VAIT is a visual analytics tool designed to analyze metropolitan transportation dataset for making sense of the data, and planning for efficient utilization. The data is acquired from vehicle Global Positioning Systems (GPS’s), and sensors deployed on the road banks. The test data in  is from 15,000 taxis running for two months in a city in China with population of 10 million. The data preprocessing step involves a novel calibration technique to account for missing or corrupt data owing to high vehicle mobility and variance in environment. Owing to the volume of the data, queries are made efficient by performing them on road segments, and then embedding taxi GPS data on the concerned road segments. To visualize trajectory of large number of taxis, the tool has been designed to use queries to reduce the data clutter, and improve caching to make visualizations responsive as well as make the system scalable. VAIT has been designed to solve 12 different predefined queries, e.g. finding at a given time, (Q1) the location or trajectory of given taxi at given time, (Q2) top n speeding taxis at a given time, (Q3) top n taxis by revenue, (Q4) origins and destinations of trips, (Q5) top n taxis by number of transactions or rides, etc.
After the execution of a novel weight-based calibration method, the visual display of movement data is captured using a novel visualization aided mining approach, called visual fingerprinting (VF). The visualizations are either trajectory or heatmap visualizations. In trajectory-style VF, the trajectory is rendered using B-splines, where the length of the trajectory gives the trip distance, and its color gives the number of visits to the destination location using average speed of each trip from source to destination. In the heatmap-style VF, one uses the index matrix constructed with number of taxi samples on the roads, average speed of these taxis, or the rank of number of passengers getting on/off at the location. The larger values imply “hotter” the map. A circle is used for encoding daily or monthly distribution of the taxis on the road.
The cartographic map is used for display of statistical information, which gives leads on interesting locations to explore further. The overview also includes the traffic flow visualization. Subsequent to the overview, the users can drill down to specific road segments or locations and do further analysis. The spatial analysis is done using the heatmap, where the spatial distributions of the features are computed. This is then used for studying temporal trends. For temporal analysis, the visual fingerprints are associated with temporal information, e.g. time taken, average speed, frequency of passing taxis, etc. The temporal analysis is done for 24-hour time period over a week or a month. The hourly distribution is colored, and temporal and spatial changes are observed by comparing different fingerprints. VAIT system relies on the querying and filtering process after the global overview, for facilitating scalability. VAIT is one of the pioneering visual analytical system which is scalable and which embeds traffic flow results on spatio-temporal movement data.
TrajGraph is a visual analytics method which is integrated with graph modeling, for studying urban mobility using taxi trajectory data. A graph is used as a data structure to store and organize information of the taxi trajectory over city streets. Then a graph partitioning algorithm is run to split to regional-level subgraphs, as opposed to street-level subgraph. The visualizations include node-link diagram of the graph without map information, a map-based visualization (for symbols and trajectories), and a temporal data visualization. These three visualizations are composite using juxtaposed views, and are linked. There is a score of importance for different regions, which is computed using Pagerank and betweenness centralities of the graph. The graph partitioning provides different scales of the data, where the multi scale visualization includes city-wide graph, then region-level, and then street-level.
The centralities are compared across time, e.g. a location with high Pagerank but low betweenness indicates a frequently visited location but a weak connection to reaching other location. A segment with high betweenness may be perceived to be a fast bypass route, and its betweenness dropping during rush hours indicates it is less preferred during the rush hours. Similar to temporal analysis, the graph modeling allows spatial analysis using the local regional analysis. One can identify high traffic regions as locations with high Pagerank and high betweenness
An urban transportation expert would use TrajGraph for troubleshooting traffic congestion. TrajGraph, a web-based prototype, opens the city-wide graph in the node-link graph view, and then shows betweenness centrality using color, for the 24-hour period. For road segments with high betweenness centrality, one can drill down further to see graph view and the map view. Given the map view, one can also drill down areas neighboring to problem areas. One can compare temporal variations in betweenness across different locations or regions, to understand their significance better. One can study variations leading to the rush hour traffic. One can identify urban traffic bottleneck, and study its pattern over a time period of time to confirm problem areas. One can then improve on the traffic congestion by resolving traffic flows during critical time-periods in these problem spots.
4.2 Broadening the Scope of Geovisual Analytics
As can be seen from Section 4.1, the design, implementation, and usage of the visual analytics solution depend heavily on the applications. At this juncture, we can add several research directions to the area of geovisual analytics. These research directions are significant in opening up research challenges and opportunities in pushing the area of geovisual analytics forward. We discuss two such research directions in this paper.
In the case study, we have discussed how introducing a data structure, such as graphs, or introducing a data preprocessing step, such as filtering, improves the visual analysis. Here, we discuss how we can integrate different domain-specific analyses which can improve on the visual analytics solution. The first research direction is on studying such amalgamations. We discuss further on one such amalgamation, namely, between spatial and social network analyses to give rise to geo-social network analysis. The second research direction is on user interfaces for geovisual analytics. An example of such a user interface is one with natural user interactions, that exploit the spatial context of the geospatio-temporal data. That would imply that one can use modes of human-computer interactions other than on a conventional personal computer, namely an immersive virtual reality.