Pattern Recognition of Movement Behavior for Intersection Classification using High Frequency GPS Trace Data
The classification of intersections (assign labels to intersections according to the type of traffic regulator) is motivated by the need for detailed and up-to-date maps. The objective of this thesis focuses on the classification of intersections based on their travel time, which is indicative of the traffic flow and the regulator that rules them (see Fig. on the right, traffic flow under traffic light and priority traffic sign).
Different Machine Learning techniques can be applied for the task of defining and recognising travel patterns along intersections. The first task of this thesis concerns the suggestion of a suitable method for this pattern recognition task (available literature exists). The second task concerns the estimation of the average travel time given the different travel patterns observed at intersections, so that these estimates further be used as features for the final categorization task, which is the labelling of intersections according to the type of regulation. A possible solution for the travel time estimate can be derived by using known supervised techniques such as linear regression.
In total under this thesis work, the student will have the opportunity to experiment with known Machine Learning techniques, using geodata (GPS vehicle trace data) and studying their applicability on a realistic scenario.
Tasks and time schedule
- Literature review on estimation of intersection travel time
- Algorithms’ implementation
- Experiments and testing
- Writing thesis documentation
Data and Tools
- Dataset: GPS trace data for training and testing of the algorithm(s)
- Tools: PostgreSQL database with PostGIS extension (tutorial of installation and usage is provided), Python and relevant packages
- Programming: knowledge of a programming language (preferably Python) as the tasks are heavily dependent on implementing algorithms
- Data Analysis/Machine Learning/Databases: prior experience is desirable but not mandatory
- Interest and willingness to work on a Machine Learning related topic
- Ability to work independently
Contact: Stefania Zourlidou