The objective of the research was to build a method which from a vast amount of movement data:



automatically identifies the used transportation modes:



Because of the amount of available data and the classification, it was possible to generate interesting plots, some not directly related to the objective of thesis.
Next is shown a plot of the classified data in the spatial extent covering the city of Amersfoort, the Netherlands. The plot contains over a million GPS points of thousands of trips made by hundreds of people. The red colour represents walking, while green are traces from bicycles, and blue from cars. The yellow colour (slightly visible in the diagonal trajectory) represents journeys made by train. Note the aggregation of the yellow points at the two train stations in the city (upper-right corner, and the other in the lower-left).



The images contain the four modes car, walking, train, and bicycle since these are the most frequently used in the Netherlands, and the inclusion of other classified modes (bus, tram, metro, etc.) is left out for aesthetic reasons. The next plot shows a larger spatial extent where the nearby city of Veenendaal is visible. The railway which runs through Amersfoort is clearly visible, as the intercity highways from the concentration of blue (car) trajectories.



The centre of Delft is mostly pedestrianised. The bicycle paths of the TU Delft campus are also clearly visible (south-east). In wider roads it is possible to distinguish the lanes.



Plotting the speed of points clearly reveals the type of roads.
The speeds are in kilometres per hour.



The same type of plot of a wider area (the cities Utrecht, Amersfoort, and Veenendaal can be recognised).



Zooming in the data is also interesting because it is possible to see the collective deceleration before roundabouts, traffic lights and similar traffic features.



At the end of the project after the whole dataset was finally classified I generated other non-geographic plots, such as the analysis of the usage of various transportation modes over time. The peaks of general movement (black) can be found at rush hours.



It was also interesting to examine the average speed of the frequent transportation modes (car, bicycle, train) over time of day. The sudden decrease at rush hours is due to traffic jams. The speeds of trains and bicycles are stable since they are not significantly affected by traffic jams.



The distribution of speeds decomposed for the frequent transportation modes shows interesting peaks for cars (values close to speed limits). Once again, the system classified between ten different modes, but here I outlined just three modes for clarity.



In frequently sampled trajectories the approximate speed at each GPS point was obtained in two ways: from the value directly calculated by the GPS receiver (thanks to the Doppler shift), and the distance between two subsequent points over their time difference. The latter is rather an average value of travel between two points, but since the sampling periods were frequent enough this data was good enough for this purpose.
The next plot shows the differences between these two values in a histogram (to me it looks as a Laplace distribution).



Now something different. Here is the spacetime plot (2D+time) of two similar journeys of the same person in two consecutive days (commuting + a small trip).




All plots had been generated with Matplotlib.
Thanks to the Section Urban and Regional Development at the OTB department for sharing the movement data that I have classified in the thesis.
Feel free to contact me if you would like to use some of the images for your needs.