Dealing with the topic of efficient computer vision, our group has a long and successful history at the chair. Our current focus lies on the technologies for driver assistance systems, which represent an important and challenging field of application. These intelligent systems analyse the vehicle´s environment via different types of sensors, for instance video and radar, thus, increasing safety and comfort for the driver.
Driver assistance systems have to provide reliable results within natural, and therefore, complex and highly dynamic environments. At the same time mobile computers suffer from limited resources. These circumstances make the given tasks very demanding.
In the presented research areas our group is able to look back on numerous projects in cooperation with companies from the automotive industry. The group has excellent know-how in computer vision using an own development framework for scene representations for the integration of a large set of modules. Particularly, in the fields of machine learning, cognitive systems and optimization we benefit from the institute´s interdisciplinary character.
In 2011, the Real-Time Computer Vision group published The German Traffic Sign Recognition Benchmark at the International Joint Conference on Neural Networks (IJCNN). The benchmark covers a single-image, multi-class classification problem, offering a large, lifelike database. The data set comprises more than 40 classes in more than 50,000 traffic images and has been used for testing and evaluation by research teams around the world.
The German Traffic Sign Recognition Benchmark
Stallkamp, J., Schlipsing, M., Salmen, J., & Igel, C. (2011). The German Traffic Sign Recognition Benchmark: A multi-class classification competition. In Proceedings of the IEEE International Joint Conference on Neural Networks (pp. 1453–1460).
In 2013, The German Traffic Sign Detection Benchmark was presented at the International Joint Conference on Neural Networks (IJCNN). It proposes a single-image detection problem. The data set includes 900 images, divided into 600 training images and 300 evaluation images, which have been sorted into one of three categories in order to suit the properties of various detection approaches with different properties. The given online evaluation system provides an immediate analysis and ranking of the submitted results. The benchmark has been used by researchers worldwide to test and evaluate their algorithms on the given task.
The German Traffic Sign Detection Benchmark
Houben, S., Stallkamp, J., Salmen, J., Schlipsing, M., & Igel, C. (2013). Detection of traffic signs in real-world images: The German Traffic Sign Detection Benchmark. In Proceedings of the IEEE International Joint Conference on Neural Networks (IJCNN) (pp. 1–8).
This tutorial was designed for students who are neither familiar with image processing nor C++. A brief introduction to both topics (in combination) is given in the .pdf file provided below. Alternatively, a .zip-folder including the .pdf, demo code and image samples can be downloaded.
(Version 2.0, 19th August 2010)