VII. PRESENTACION DE ANALISIS DE RESULTADOS
7.3 Emociones que activan los estudiantes como respuesta a las expresiones emitidas por
Figure 2.3.1: Discontinuities in the time-stamp of AIS position reports belonging to a single MMSI aggregated time-series
geographic region and provided a time ordered stream of position reports from a single MMSI number. The stream of position reports might contain time-series of multiple voyages at different times since the data is aggregated based on MMSI number and an example of this is seen in Figure 2.3.1 where the time-stamps from a single MMSI number data stream is shown. These time-stamp discontinuities are a result of a vessel leaving and reentering an area, and the “contiguous” sections of time-stamps represent single voyages which can be separated from the data stream based on the time-stamp differential.
A single MMSI aggregation may also contain a data stream of a vessel which enter the area and proceeds to a harbor. These data streams can be refined into single voyages by considering not only the continuity of the time-stamp but also the magnitude of the vessels speed over ground since a stationary vessel should not be included in the statistical base of traffic patterns and neither should the continuous stream of position reports with stops at a harbor give rise to a single traffic pattern, but to two patterns entering and leaving the harbor. A single voyage can be separated from such a stream if the speed of the vessel drops below a defined threshold. The combination of these procedure results in a set of time-series of dynamic AIS data which each represent a single voyage which can be used to build a statistical description of the traffic patterns.
2.4
Single voyages to aggregated traffic patterns
While the accuracy and sample rate of data collected from AIS is of sufficient fidelity for analysis, data from a small selection of voyages will not reveal the predominant traf- fic patterns in an area and will not provide a dataset of sufficient size to derive adequate statistics. A large number of data series must be collected during a long interval which is trivial with AIS as long as the data is stored for later use. The challenge when working with AIS to analyze traffic is not to acquire the data, but analyze it efficiently in order
to take advantage of the large number of observations. AIS is a continuously observing system and will during a three month period with a small number of passages (10 pr day) generate in the order of 1000 position series. Manual processing of data-sets of this mag- nitude quickly become infeasible, especially considering 10 passages pr day represents a very modest density. An automatic method for aggregating individual time-series into groups of similar voyages is an prerequisite in order to fully exploit the AIS data. A method of converting individual dynamic AIS messages to data streams from individual vessels was presented in the previous section, but only represents a transformation from disjoint observations into a base for analysis.
Since the simple traffic model is an idealized geometric structure the analysis of vessel traffic must begin with an separation of the geometric traffic patterns present and associ- ating the vessel tracks with the different groups. Aggregating time-series from AIS based on the path taken through an area is essentially a problem of pattern recognition. If the task was to be completed manually the decisions about which time-series formed a group of similar traffic would be based on a subjective measure of similarity where the operator would assess the geometric properties of two specimens and classifying them based on the question “are they similar?”. Automatic comparison of the similarity of objects has long been studied in computer vision in order to have a computer algorithm separate the contents of an image, determine the features of the objects present and to decide if the same object is present in several images. The results produced by these algorithms can be seen in satellite image mapping programs where they are used to stitch together over- lapping aerial photos. The body of knowledge represented by the developed algorithms for image processing is a valuable resource which can be applied to other problems, if the problem lends itself to an image representation where a bitmap image is composed of one or more matrices of color intensities.
Before one can start to analyze the data streams one must decide what to analyze, in this case the obvious answer is to analyze each individual data stream based on the time-series of the position which leaves a trace through the area. Transfer of a time-series of posi- tion data into an image representation can be achieved by discrertizising the area (matrix formulation) and counting the number of position reports in each discrete region (color value). This transfer of remotely sensed data into a known representation achieves addi- tional benefits since the camera position and exposure/capture conditions for the image is controlled by the application. This simplifies further analysis as the problems of different exposure, lens distortion and camera orientation can safely assumed to be absent.
The first attempt at aggregating vessel tracks by image processing from (Aarsæther and Moan, 2008) was based on comparing exaggerated, a reduction in resolution, track-lines in place showed promising results. The direct comparison procedure is illustrated in Fig- ure 2.4.1 where both the original and exaggerated representations and the difference be- tween them are shown. The figure shows that a successful match of two vessel tracks can be expected if the track placed the vessels at similar positions along the track, how- ever the comparison will fail if the tracks are on opposite extremes of an imagined mean track-line. This effect can be suppressed by extending the exaggeration of the track, but
2.5. TRAFFIC PATTERN ANALYSIS 29