Master's Thesis presentation, Emma Darebro

Abstract: This thesis introduces a modern way of estimating the risk for a vehicle to do a lane change. Autonomous driving features in vehicles are increasing rapidly, and with that, the interest in finding risk predictions for actions is vital. Today the risk factors specific to lane changes are mainly based on machine learning trained on annotated data collected from external sources. This thesis investigates the same approach but on data collected from radar and camera equipment attached to the vehicle performing the lane change. We raised the question \textit{Is it possible to predict a lane changes risk of causing an accident during the lane change performance}?. To answer this question, the tree-based ensemble model \textit{Random Forest} was used to create several models. All used different sized data sets representing different stages in the lane change. The conclusion drawn is that it is possible to accurately predict a lane changes risk halfway in the lane change performance with high precision. Concern regarding the data set was raised, and it was deemed that it contained too little variation to be seen as sufficient.

Password: 415748