Odometry Evaluation
Yukyung Choi (KAIST), Namil Kim (NAVER LABS), Kibaek Park (KAIST), Soonmin Hwang (KAIST), Jae Shin Yoon (Minessota) and In So Kweon (KAIST)
Dataset Information
GPS Trajectory for one sequence. Here we plot the GPS overlaid on top of an aerial image from Google maps. Colors encode the GPS signal quality: blue tracks have been recorded with high precision using RTK corrections, green denotes the general GPS information without correction signals. The red indicates the shadow area which one have been excluded from our data set. W, E and N indicate each sequence in our datasets.
- download RGB-Thermal Pairs
- download baseline code
Evaluation Results
We present Extended DLoop as our baseline algorithm to verify the compatibility of previous algorithms and handle the visible and thermal images. Compared with a single-spectral result, our extended baseline acceptably shows better performances in all-times. This implies that the complementary information of visible and thermal image is helpful regardless of any illumination conditions.
GPS positions of the images are plotted with a black dot. Wherever an image matches a loop with another image, the position is labelled with a red dot. The precision is 100% in all cases. AM04, AM06 and AM11 sets are used to train a vocabulary tree.
