search

VIS4ROB-lab / ccm_slam

CCM-SLAM: Robust and Efficient Centralized Collaborative Monocular SLAM for Robotic Teams
540 stars 107 forks source link

run on the kitti dataset #5

Closed ghost closed 5 years ago

ghost commented 5 years ago

Hi, when I ran this code on the kitti dataset, it always failed to initialize. And I had adjust conf/vi_euroc.yaml correctly.

patriksc commented 5 years ago

We ran a brief test on KITTI (sequences 00,01,02,03). The motion pattern of the car is not optimal for monocular system such as CCM-SLAM, since it is mostly movement in the camera axis, causing problems with initialization and drift. We converted the single image files to a rosbag files for our tests. On seq. 00, CCM-SLAM initializes for us. We also tried to with down-sampled images (factor 0.5), and with this data, also seq. 02 and 03 initialize after approx. 23 and 15 sec respectively, this seems to alleviate the initialization problems.

soulslicer commented 4 years ago

@patriksc How do you run it on the KITTI dataset? How did you convert the single image files into rosbag files. What topics are required for ccmslam? Can you provide some KITTI bag examples?

patriksc commented 4 years ago

Unfortunately, the bagfiles are quite large, so we cannot provide them here. However, converting images to a bagfile is quite straightforward, you can for example use or take inspiration from these tools:

https://github.com/ethz-asl/kitti_to_rosbag https://github.com/raulmur/BagFromImages

Maybe you want to make some to the code adjustments, for example to change the resolution of the images.