Traffic accidents are notoriously difficult to predict. This is because of the many factors involved during the accident(like driver reacting and braking on time or the presence of other vehicles during the accident). Previous accident analysis rely on information from accident data bases. However, these databases usually do not contain all the information about the accident(for example, the cause of the accident ,all relevant participants during the accident). Secondly accident do not happen very often, so the available data is limited and often very sparse in time and space. One alternative to accident database is the use of traffic cameras to capture in real-time traffic interactions. These captured videos can then be processed using state-of-the-art video imaging techniques to extract trajectories of vehicles. These trajectories often contain traffic conflicts which did not lead to accidents(near misses) and can be used(in theory) to estimate the expected number the accidents. This project will develop a methodology to estimate the expected number of accidents at an intersection using the number of extracted traffic conflicts from that intersection as an input. The aim of the project is to be able to use video footage of a few days of traffic monitoring at an intersection to estimate the number of accidents for that intersection.
Research objective
The assignment will involve a literature study, looking into organizations that have video data processing of trajectories extracted from video data from an intersection and finally the development of the methodology. The focus will be on the methodology to translate number of conflicts to number of accidents using probability of collision of a conflict as input. A knowledge of python programming language is required. There is also a conflict simulator developed (in
VISSIM) which can be used for simulating a conflict and estimating a collision probability. This could be used as input to the method but for now these probabilities can be considered given. A knowledge of traffic simulation is desired but not required. For the development of the method, a good background in mathematics, statistics, machine learning is highly recommended.
Approach
- Perform a literature study of available methods to translate number of traffic conflicts to number of accidents.
- Develop a methodology to estimate the expected number of accidents at an intersection using the number of extracted traffic conflicts ( and their collision probabilities) from that intersection.
- Test and Validate the methodology using real video data from an intersection.
Supervision
Are you interested in this assignment? Contact the Master thesis coordinator:
