LiDAR Object Detection and - Sensor Fusion in Simulation Environments Sensor modelling towards advancements in Real2Sim - Sim2Real

Abstract

In the world of autonomous driving the environment perception is of the outmost importance. Light detection and ranging(LiDAR) data is one of the most common sensor data in the field of autonomous driving. This thesis will therefore explore LiDAR data as the first steps toward using it in a simulation environment. In order to be able to explore the use of LiDAR, an arsenal of tools and knowledge is needed. This thesis goes through the learning steps of these tools and in which way they fit together to arrive at the result. This thesis includes different approaches of visualizing data using Robotic Operation Systems (ROS) framework and implementing 3D bounding box annotation and also covers creating a convolutional auto encoder and implementing it as a deep neural network. The result is knowledge on how LiDAR data works, how it can be used and estimating some characteristics using a deep neural network. Included in the results are a few visualizations of the data after being trained on a neural network. Further work once this knowledge is gathered would be to implement it into a simulation environment.