Image Feature Usability Testing in Various Environments
Name
Heiti Ehrpais
Abstract
In this thesis, a method for evaluating the performance of different feature descriptors for structure from motion (SfM) algorithms is presented. This method is used to compare the performance and time consumption of different configurations of Scale-invariant feature transform (SIFT) and Accelerated KAZE (AKAZE) feature descriptors in various off-road environments. This is achieved by performing SfM based visual reconstruction on data gathered by an unmanned ground vehicle (UGV) and comparing the quality of the results. Performance of the algorithms is compared by evaluating the positional and rotational correspondences between two co-localized cameras. The visual reconstruction performs well in all chosen conditions, which include the forest, open field, sand and different illumination conditions. The different feature extractors and their parameters have an impact on the time consumption, reconstruction success rate and the accuracy of reconstruction. The method to compare the feature descriptors and results from this comparison is presented in the thesis, along with the results obtained by using it.
Graduation Thesis language
English
Graduation Thesis type
Bachelor - Computer Science
Supervisor(s)
Mihkel Pajusalu
Defence year
2020