Back to Top

Trajectories Analysis

One of the main goals of the VaVeL project is to identify and address citizens’ needs and improve their quality of life. A major source of information that demonstrates the citizens’ needs is the set of location data that are traced while they move in the city. More specifically large amounts of trajectory data are generated and can be available either from citizens that allow to disclose their trajectory information, traced from their mobile devices, or from means of public transportation like buses or trams. The VaVeL Use Cases consider real time location data generated from buses moving at the city of Dublin and buses and trams from Warsaw.  

The analysis of trajectory data will provide beneficial information to the city stakeholders providing them with valuable information regarding the understanding the citizens’ movement in the city, the real time detection of anomalous traffic events at different locations of the city and the detection of the current traffic condition at the road network. Nevertheless, trajectory data, like other urban data suffer from several issues like noisiness, sparsity, veracity and difficulty to understand, which severely reduces their usability. We have focused on three main problems that appear in the analysis of trajectory data, that are described bellow:  

  • Trajectories Sparsity:
    The collected GPS data are usually retrieved from mobile devices installed on moving vehicles. Frequently, these devices run on limited bandwidth and power resulting to sparse trajectories, of varying sampling rate, making difficult their interpretation. This fact, poses challenges for understanding the exact path the vehicle follows. Urban planing, traffic management and location based services are hindered by the uncertainty that is generated due to the trajectories’ sparsity. Thus transforming sparse to dense trajectories is required for their effective analysis and processing. An example is illustrated on Figure 1, where the GPS locations reported by a moving vehicle are shown, generating uncertainty regarding the road that the vehicle followed. Clearly, the path followed the vehicle in order to transit from t 3 to t 4 is ambiguous. Motivated by this problem, in D4.2, Section 2.1 and in [LLGG16] we propose a technique that is able to infer the complete geometry of a trajectory from sparse GPS points.
    Ενσωματωμένη εικόνα 1
    Figure 1: GPS samples reported from a vehicle moving in the city of Dublin. The sparsity of the measurements poses challenges in inferring the actual path that is followed.aption
  • Maps’ Veracity:
    Digital maps are very useful for providing a variety of services to the citizens. A citizen may schedule his trips and his daily routine according to a digital map. At the same time, public transportation services often rely on digital maps in order to plan the transportation routes or in order to dynamically alter the routes in the case of an unexpected event. Clearly, in order to provide such services, access to an up-to-date map is required. Unfortunately, access on an up-to-date map is not guaranteed and often very challenging to obtain. Figure 2 shows an example taken from OpenStreetMaps that depicts two map regions that are inconsistent with the actual map network. On the top figure, the network available from OpenStreetMap is illustrated while on the bottom figure the satellite image for the same period is presented. In order to handle such cases, in
    D4.2, Section 2.2 and in [CLH+16] we describe a framework capable of automatically generating and updating digital maps using large trajectory collections.    
    Ενσωματωμένη εικόνα 2
    Figure 2: Identified regions where OpenStreetMap does not agree with satellite images of the same period. Notably, novel automatic map enrichment techniques are able to automatically update the map. The satellite image aligned with GPS points are shown at the bottom ([CLH + 16]).

  • Understanding Movement Patterns:
    The pervasiveness of mobile devices that trace the GPS location of the moving objects generates large collections of spatiotemporal data. It is particularly challenging to process these data and derive meaningful information from them. Imagine for instance, two different GPS trajectories that follow the same route at the same time, due to the different sampling rates and the noisy GPS measurements it highly unlikely to be the same. Nonetheless the trajectories generated in urban environments usually share common paths. A key challenge is to detect such common paths that are frequently followed from a considerable number of trajectories. As an example, consider the 4 trajectories visualized in Figure 3. It can be observed that even if they have different origins and destinations they share a common behaviour at the marked area. Furthermore a common path that describes the objects movement at this area can be extracted, aggregating the information of how objects moved at the marked area. In Section 2.3 of D4.2 we introduce a method that is able to detect such frequently followed paths.
    Ενσωματωμένη εικόνα 3
    Figure 3: Example of 4 trajectories, sharing a common path.


[LLGG16] Y. Li, Y. Li, D. Gunopulos, and L. Guibas. Knowledge-based trajectory completion from sparse gps samples. InProceedings of the 24th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems,page 33. ACM, 2016.
[CLH+16] C. Chen, C. Lu, Q. Huang, Q. Yang, D. Gunopulos, and L. Guibas. City-scalemap creation and  updating using gps collections. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1465–1474. ACM, 2016.