Technical Reports in Computing Science – Schriftenreihe Informatik der HS Kempten
July 2021
This paper is a summary of the creation and data usage of an autonomous model vehicle which was recreated in CarMaker[1]. The simulation data come to use when following students of this semester project will develop algorithms and simulations which are impractical to test or train in real life. This paper starts with a summary of the background information around the model vehicle and the software CarMaker[1], then reproduces the construction process, digs deeper in ROS[6] and finishes with the extraction of the data.
July 2022
This paper is a summary oft he traffic light detection for an autonomous model vehicle which was created for a student semester project. The traffic light detection comes to use when the model vehicle takes part in the VDI-Cup or other future Cups. The VDI-Cup will be discribed later on. The structure of this paper start with a summary, followed by the discribtion oft the VDI-Cup. After that we will explain how we implemented the traffic light detection and finishes with the conclusion of our implementation.
July 2021
July 2021
A real time object detection system is an essential core feature of any autonomous driving car. With the variety of models and data sets available, there is great opportunity to train a model that meets the high requirements of autonomous vehicle. In this paper we want to present an object detection model for traffic signs and lights implemented on a NVIDIA Xavier board [7] using the robot operating system (ROS) [9]. Furthermore we explain how to create and evaluate a data set as well as to test the accuracy and performance of the implemented detector. In regards to pattern recognition with neural networks, the performance and accuracy must also be weighed up. The goal is to achieve one without sacrificing the other. Considering all these aspects, we decided to train a YOLOv4 Standard and a YOLOv4 Tiny configuration [2]. While the accuracy of the YOLOv4 network exhibited a very high accuracy rate on high and mediocre resolutions but the performance is too poor to be used in a real time object detection system. On the other hand, the YOLOv4 tiny network reached the strived for performance even on the highest resolution scales tested, but at the cost of accuracy.
CS-02-2020
In the last decade autonomous driving has evolved from a science fictional dream to an everyday reality. With the advance of more and more companies bringing their versions of self-driving cars on the street it is just a matter of time before the majority of transportation will be in the hand of computers. But with the deadly car accident involving a self-driving Uber car back in 2018 there is also the question about how reliable autonomous driving really is and how we can validate and test the safety of this new road user 1. An uprising approach towards creating robust and adaptable neural networks is called domain randomization. This paper explores the possibility of using this method to create training data with driving simulations. It will propose a list of important criteria and factors affecting the selection of a fitting simulation. Furthermore it will present a track generator which is able to create useful tracks and export them to a format which can be used by several common simulations used in the field of autonomous driving research.
CS-01-2020
Collecting the training set required for building a robust neural network for autonomous driving requires large amount of data. It is nearly impossible to collect this data only by recording the driving of real world vehicles. There is no organization or company that is able to provide the resources needed to tackle this task.[1] Therefore the approach currently used is to generate the large amount of training data by simulating virtual cars in computer simulations that try to mirror real world road traffic as close as possible. Besides the huge amount it is also important that the data generated varies quiet a lot, otherwise the neural network can not learn to adapt to the many different situations that occur in real world road traffic every day.[2] Therefore it would be great to record the virtual car driving on as many different tracks as possible. To solve this issue this paper proposes a fast and simple iterative algorithm that can be used to procedural generate tracks that can be used for the recording and training of autonomous driving cars.
July 2020
Lane detection is an essential part for an autonomous car to function. With a lane departure warning system many modern cars are already equipped with a lane detection system. But models and methods to predict lanes are numerous and often follow different approaches to solve the lane detection problem. This paper wants to give an overview on some of the state of the art models and methods for lane detection that lately achieved good results on public datasets. The paper will also look at training and testing a SCNN [10] using the free Colaboratory service from Google [29].
July 2020
This paper is a summary of the time to collision (TTC) calculation for an autonomous model vehicle which was created for a student semester project. The TTC comes to use when the model vehicle takes part in the Carolo-and VDI-Cup which will be described later on. The structure of this paper starts with a summary of the background information, then digs deeper into what algorithms are discovered already and how they work and finishes with the explanation and conclusion of our implementation.
July 2020