Image based lane keeping by end-to-end machine learning within simulated environment
Başlık çevirisi mevcut değil.
- Tez No: 792387
- Danışmanlar: DR. ARADİ SZİLARD
- Tez Türü: Yüksek Lisans
- Konular: Mühendislik Bilimleri, Engineering Sciences
- Anahtar Kelimeler: Autonomous driving, Supervised Learning, Behavioural Cloning
- Yıl: 2020
- Dil: İngilizce
- Üniversite: Budapest University of Technology and Economics
- Enstitü: Yurtdışı Enstitü
- Ana Bilim Dalı: Belirtilmemiş.
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 76
Özet
Özet yok.
Özet (Çeviri)
Autonomous driving has been a popular area of research for many years. Most of studies started around 1980s with using automotive sensors like radar, lidar, cameras and others for autonomous driving tasks. Since cameras are cheaper than other sensors and improvement of machine and deep learning methods increased computer vision applications. Machine learning solutions are used in several autonomous vehicle subsystems, as sensor fusion, perception, localization and mapping, and path planning. Convolutional neural networks have enabled a new generation of autonomous vehicles and even public datasets can be used to build networks capable of predicting steering angles and detect objects for self-driving vehicles. In this thesis, end-to-end autonomous driving is demonstrated in a simulation environment by commanding steering control inputs from raw images. Supervised learning is used to map images to steering inputs based on example data from driver. 3 different convolutional neural networks are trained to predict steering wheel angles using a dataset of front view images of car. Trainings will include many experiments by changing training parameters for each model. Effects of hyperparameters on training performance will be discussed and steering outputs will be compared. Deep learning approach has the disadvantage that if a situation becomes more complex, due to the high information density of the image, enormous amounts of data are required. Furthermore, they cannot be trained in the real world or only with very high effort because the vehicle will learn to avoid collisions with objects by driving against them. Besides, the testing of functions and especially safety-relevant ones is increasing difficulty. Therefore, 3D simulation environments are required to supply a secure training and learning environment with the possibility to generate the maximum amount of data as the algorithm will need to learn realistic driving behaviour. IPG Carmaker is chosen as simulation environment which is a real time simulation software provides detailed driving conditions. By the help of Matlab interface, network is trained in Matlab and steering output is fed to simulation software.
Benzer Tezler
- Gerçek zamanlı şerit takip desteği sisteminin geliştirilmesi
Development of a real time lane keeping assistance system
ÖZGÜR TUNÇER
Yüksek Lisans
Türkçe
2010
Mühendislik Bilimleriİstanbul Teknik ÜniversitesiMekatronik Mühendisliği Ana Bilim Dalı
PROF. DR. LEVENT GÜVENÇ
- Sensor fusion of a camera and 2D LIDAR for lane detection and tracking
Şerit tespiti ve takibi için kamera ve 2D LIDAR sensör füzyonu
YASİN YENİAYDIN
Yüksek Lisans
İngilizce
2019
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolOrta Doğu Teknik ÜniversitesiElektrik-Elektronik Mühendisliği Ana Bilim Dalı
PROF. DR. KLAUS VERNER SCHMİDT
- Vision based road slope estimation
Kamera görüntüleri ile yol eğimlerinin bulunması
ESER ÜSTÜNEL
Doktora
İngilizce
2021
Elektrik ve Elektronik MühendisliğiYeditepe ÜniversitesiElektrik-Elektronik Mühendisliği Ana Bilim Dalı
DOÇ. DR. ENGİN MAŞAZADE
- Gömülü platformlar için görüntü işleme temelli gerçek zamanlı şerit tespit ve uyarı yöntemleri
Image processing based real-time lane detection and warning methods for embedded platforms
AYHAN KÜÇÜKMANİSA
Doktora
Türkçe
2018
Elektrik ve Elektronik MühendisliğiKocaeli ÜniversitesiElektronik ve Haberleşme Mühendisliği Ana Bilim Dalı
PROF. DR. OĞUZHAN URHAN
- Mobil robot platformu üzerinde DSP tabanlı gerçek zamanlı çalışan şerit tespiti algoritmasının geliştirilmesi ve optimizasyonu
Development and optimization of DSP (Digital Signal Processor) based real time lane detection algorithm on a mobile robot platform
GÜRKAN KÜÇÜKYILDIZ
Yüksek Lisans
Türkçe
2012
Mekatronik MühendisliğiKocaeli ÜniversitesiMekatronik Mühendisliği Ana Bilim Dalı
DOÇ. DR. HASAN OCAK