Inhalt
Inhalt |
E-Learning Contents, start open
Overview of the content:
1. Brief introduction
2. The pinhole camera model
3. Recognition
4. Motion analysis
5. 3D reconstruction
We will focus on automotive applications and take a look at both, traditional and machine learning algorithms. To evaluate some of these, algorithms, we will take a look at different datasets (MS COCO, KITTI, ApolloScpae, etc). These datasets can also be used as a benchmark for our projects.
For traditional algortihms, we will work with OpenCV. For machine learning algorithms, we will take a look at the TensorFlow Object Detection API.
As part of this course, you will implement or evaluate one of these algorithms using C++ or Python. The algorithms will either have to work on already recorded data (like the datasets above) or a given sensor. |
Literatur |
R. Szeliski: Computer Vision: Algorithms and Applications
http://szeliski.org/Book/
OpenCV tutorials (C++, Python)
https://docs.opencv.org/trunk/
Tensorflow Object Detection API
https://github.com/tensorflow/models/tree/master/research/object_detection |
Lernziele |
After attending the lecture, the participants will be able to understand the most important algorithms for three main tasks in computer vision:
1. Recognition: object detection, pose estimation, etc.
2. Motion analysis: egomotion, optical flow, etc.
3. 3D reconstruction: localization, mapping, using mono camera and stereo vision
At the end of the semester, the participants will implement or evaluate one of these algorithms using C++ or Python. |
Voraussetzungen |
Good understanding of mathematics in general.
Good understanding of at least one programming language, preferable Python or C++.
Depending on your project: additional knowledge and first experiences with machine learning using TensorFlow or comparable frameworks. |
Leistungsnachweis |
Siehe Modulhandbuch, Aushänge beachten
PF, wenn online PA benotet
|