Weather effects on short-range LiDAR and their classification

Abstract: Today we are seeing exciting developments in the field of autonomous vehicles, on both software and hardware. Veoneer is a company making a contribution where research and manufacturing is being done on hardware and active safety. One of the most important aspects in this field is road safety, where understanding the behaviour of sensors used in vehicles is essential. From the point of view of safety, understanding how weather affects the sensors is necessary for a successful deployment. This study is a continuation of previous studies done at Veoneer, and regards how various adverse condition affect the performance of a short-range LiDAR and gives a thorough description of the involved physical processes. Data collected over a couple of months was analysed and compared to theoretical models in order to establish their validity. In addition, LiDAR measurement were done in a chamber where conditions could be varied in a controlled manner. Furthermore, analysis methods were used to transform the data into a form potentially more useful for use in machine learning algorithms to estimate the ability to classify conditions based on LiDAR signals. The used models showed mixed results, with some showing more agreement than others. Models regarding foggy conditions generally showed greater agreement with data than in other conditions, although some variation around the predictions did occur. In regards to the performance of the classification algorithms, there were als omixed results, where the sensitivity in fog was at most 96 % and the precision at most 64 %. This thesis also enables and suggests further research into the utility of short-range LiDAR both in the field of autonomous vehicle safety as well as in use of other fields such as meteorology.