|Title:||Deep learning to classify driving events using GPS data|
The purpose of this thesis is to develop a model capable of classifying driving events using segments of GPS sensory data. The work to develop the model will be conducted based on semi-supervised/ unsupervised machine learning approach.
The model is expected to segment and detect a given set of events like roundabouts of all shapes, left and right intersection turns using GPS sensory data that correspond to a vehicle path. However, involving more events like overtaking or identifying additional events in an unsupervised approach is highly encouraged.
We propose to employ deep learning algorithms to classify GPS path segments based on map images. However, the students are encouraged to investigate other artificial intelligence algorithms to resolve the research problem.
1. Literature study and state-of-the-art
This task requires a systematic literature review to identify relevant features, algorithms, metrics and approaches of classifying driving events using GPS path segments that correspond to a vehicle path. Student requires presenting an analytical summary of the state-of-the-art based on the literature study.
This task involves analyzing the Pedestrians' data using IMU sensor and GPS signals of various walking/running events and developing an approach using machine-learning algorithms for detection of these patterns based events.
Student should evaluate the proposed approach and learning algorithm for detecting driving events. A statistical analysis of the result shall be provided to prove the significance of the results.
4. Report writing
It is expected that student provide a report as a completion of the project work. Report consists of background, problem formulation, state-of-the-art, methods, evaluation, and discussion.
5. Regular Supervision Meeting
Regular meeting shall be planned and conducted on a bi-weekly basis (frequency of one meeting every two weeks) to allow for constructive discussions and control of the work progress and research deliverables.
|IDT supervisors:||Mobyen Uddin Ahmed, Catharina Bexander|