Behaviour modeling and classification of vehicles at a roundabout
| Title | Behaviour modeling and classification of vehicles at a roundabout |
|---|---|
| Summary | Modeling of behaviour, classification based on behaviour, and detection of anomalous behaviour in traffic at a roundabout. |
| Keywords | Behaviour modeling, lidar |
| TimeFrame | Oct 2018 to June 2019, with possible extension to Sep 2019 |
| References | |
| Prerequisites | Programming (any of C++, Python, Matlab) |
| Author | |
| Supervisor | Björn Åstrand, Naveed Muhammad |
| Level | Master |
| Status | Open |
Context: This master thesis falls in the context of autonomous driving and more specifically infrastructural support for autonomous driving. An autonomous vehicle needs to understand and act according to the way other vehicles around it are behaving, for the purposes of safety and efficiency. This project focuses on modeling the behaviour on vehicles at a roundabout.
Data: The dataset used in the project is multi-beam lidar data gathered at a busy roundabout. This dataset is in the form of raw lidar perception data saved as ROS bag files. Processing of data (segmentation and extraction of vehicle trajectories) are also a part of this project. It might also be possible to acquire/use more data from analogous scenarios.
Research questions: The following research questions will be addresses in the project.
(i) Which machine-learning approaches are suitable for behaviour modeling of vehicles at a roundabout? (ii) How can vehicles in traffic be classified into different categories, based on behaviour? (iii) How can anomalous behaviours be identified in a roundabout scenario? (iv) Can a parametric model be developed which can also act as a supervisor to accept or reject an intention estimate of a vehicle generated by a behaviour modeling layer?
Work packages: (i) literature review, (ii) segmentation and trajectory extraction from lidar data, (iii) developing a machine learning approach for modeling, classification and anomaly detection, in terms of vehicle behaviour, (iv) developing a supervisory layer to validate agent intentions (v) [bonus] conference publication.
