Chair of
Multimedia Communications and Signal Processing
Prof. Dr.-Ing. André Kaup

A Comparison of Feature Detectors and Description for Tracking of 2D Navigation Maps

Supervisor:Dr.-Ing. Gilbert Yammine
Faculty:Prof. Dr.-Ing. André Kaup
Student:Ali Khairat
Begin:2011-06-15
End:2011-12-15
File:These-PDF
Info:Quality monitoring and error detection in systems is a major step in the




testing phase of a system before making it available to the consumer.




In new infotainment systems present in luxurious cars, system errors




are not accepted and should be avoided. Consequently, a monitoring




system should be built to record the data shown on the infotainment




screen and an analysis of the data should be done to detect errors and




find their sources.




One frequent system error is image freezing, i.e. motion stop of the navigation




map due to high CPU loads or other error sources. In order to




detect such an artifact, the difference image between frames could be




used to signal any motion in the frames. When no motion exists, this




can mean that either the car has stopped or a freezing has occurred. In




order to differentiate between both cases, the map position should be




calculated before and after the motion stop, and map jumps could then




be detected. For that, invariant points on the map that exist before and




after the freeze should be found and map rotations and translations




could then be calculated by solving the affine transformation equation.




The task of Mr. Khairat is to test different invariant feature detectors




and descriptors such as Harris, SIFT, SURF, and FAST, and compare




their performance when used for tracking of translating and rotating navigation




maps considering different map display styles. Two performance




criteria have to be tested: the robustness of the features after rotation




and translation of the map, and the computation time for each detector/




descriptor. Additionally, Mr. Khairat has to implement a method




to select the best feature points and track the motion of the map. He




will have then to test his algorithm on different navigation sequences




(with and without freezing) and report its performance for various map




jumps.




The algorithms have to be implemented in MATLAB and a clean documentation




of the work and of the source code will be highly assessed.




Betreuer: Gilbert
TypeMaster Thesis
Status:Terminated