CEP-traj: An event-based solution to process trajectory data

Highlights

• A novel palette of datasets for CEP-traj and VAbDS evaluation.

• Evaluation of the middleware with network-constrained trajectories.

• Comparative of CEP-traj and a well-established online trajectory-processing framework.

• Evaluation of VAbDS by using a public real-world dataset.

Abstract

In this day and age, there exists an increasing need for systems and architectures able to process spatio-temporal data in a timely way. As a result, this paper presents CEP-traj, a novel middleware to ease the development of real-time trajectory-based services based on the Complex Event Processing (CEP) paradigm. By means of an event-based approach, the present middleware is able to detect a set of generic patterns along with meaningful changes of an entity׳s movement. In order to prove its suitability and feasibility, a vessel abnormal-behaviour detection system has been developed on the basis of the middleware׳s features. Finally, both synthetic and real datasets have been used to test the accuracy and performance of the middleware and the detection system implemented on top of the Esper engine.

Introduction

The increasing pervasiveness of location-acquisition technologies has enabled the collection of a huge amount of trajectories for a varied range of moving entities. This has led to an increasing interest in the development of trajectory-based services in quite different fields. For that reason, many off-line solutions to analyse trajectory data have arisen for the last years [46]. Nevertheless, there exists a pressing need for detecting new and relevant trajectory knowledge in nearly real time so as to achieve fast reaction times. This has motivated the recent development of on-line approaches to perform different trajectory processing such as segmentation [18], [42] or annotation [42].

Regarding the detection of trajectory-based patterns, most of the current on-line solutions focus on moving groups by means of clustering-based approaches [38], [40], [35]. However, other finer-grained patterns related to the interactions or relationships that may occur between pairs of moving objects are a topic that requires further attention. These motion-based relationships may be of great importance in several fields like behavioural ecology or threat detection. As a matter of fact, a wildlife monitoring system could be interested in detecting whether some type of predator is actually moving towards the area where a protected animal is currently resting. Also, a vessel traffic service would be interested in perceiving as soon as possible whether a vessel is approaching to another as this might be a sign of an illegal rendezvous. In these two cases, the situation of interest only involves two particular entities (the predator and the protected animal in the first example, and the two vessels in the second one) rather than a large group of moving entities.

In order to detect patterns in time-constrained domains, the Complex Event Processing (CEP) model has emerged as a suitable solution [13]. CEP is a novel paradigm that focuses on timely processing unbounded flows of information items, so-called events, from a great number of distributed sources to detect certain activities of interest following a push-style communication.

Taking into account that a moving entity׳s trajectory is potentially infinite, dynamically changing and must be processed in almost real time, the present work rests on the hypothesis that CEP is a suitable approach to process and analyse the spatio-temporal data of a moving entity. As a result, CEP-traj, a general-purpose CEP-based middleware to timely analyse trajectory data, has been developed.

CEP-traj basically processes raw traces of timestamped locations defining the trajectories of a set of target moving objects or entities (people, animals, vessels and the like) as soon as it receives them. Next, on the basis of these traces, it infers a set of meaningful situations in the form of derived events that are asynchronously delivered to a back-end service. In particular, CEP-traj processes the incoming locations by means of a set of chained steps, data cleaning, data segmentation & compression and pattern detection. Each of these steps has been developed as lightweight mechanisms that only require a single scan of the incoming data. This has been done by means of abstract operators well-established in the CEP core. This allows to eventually deploy the middleware on top of a varied range of commercial CEP commodities.

Concerning the pattern detection step, the proposed middleware is capable of timely discovering a set of motion-based generic patterns. Specifically, CEP-traj distinguishes two types of patterns, (1) the patterns that might exist between pairs of entities such as converge or diverge, and (2) the ones between a moving entity and a stationary geographic area. The detection of these patterns is a powerful tool in order to develop certain types of services as it has been mentioned above.

Furthermore, in order to prove the suitability and feasibility of the proposal, the present paper also puts forward a Vessel Abnormal-behaviour Detection System (VAbDS) which has been entirely developed on the basis of CEP-traj. The goal of this system is to detect several behavioural patterns among the vessels, which give insight into certain illegal activities, on the basis of the generic patterns provided by CEP-traj.

To sum up, the salient contribution of the present work is to introduce a novel event-based approach for on-line trajectory analysis. Despite the fact that other online middlewares also comprising trajectory cleaning, segmentation and compression phases have been already proposed, they centre on the timely annotation of the incoming trajectories where no pattern discovery is considered [42]. The present work also differs from previous online pattern-discovery proposals [38], [35], [24] as it focuses on a varied range of motion-based patterns involving pairs of moving entities instead of group-based patterns. This has implied developing incremental mechanisms to clean, partition and compress the incoming trajectories in near real time.

The remainder of the paper is structured as follows, an overview of the state of the art of the trajectory analysis and the CEP domains is put forward in Section 2. A detailed explanation of the CEP-traj middleware following an event-processing approach is stated in Section 3. Section 4 is devoted to describing the VAbDS. Then, Section 5 discusses the results of the different experiments. Finally, the main conclusions and the future work are summed up in Section 6.