Master's Thesis presentation, Jesper Jäghagen and Carl Hjalmarsson

Abstract: The aim of this Master’s thesis project is to develop deep learning models which generate a two dimensional occupancy grid around a test vehicle. To achieve this, ultrasonic sensor (USS) data gathered by the test vehicle moving in low speed environments is used as input. The goal of the project is to make the resulting occupancy grid mimic the grid created from the more detailed short range reference system Lidars and to learn any hidden patterns. The thesis objective is approached by first exploring the available USS and Lidar data streams and defining proper representations of the data. The data is then processed to acquire the desired representations and deep learning models suited for the problem are presented. Finally, these models are trained and there after evaluated with five different metrics to asses how well they perform compared to each other and benchmark evaluations. The outcomes and approaches are then discussed to draw final conclusions as well as suggest future work.

Examiner: Adam Andersson