search

HKUST-Aerial-Robotics / D2SLAM

$D^2$SLAM: Decentralized and Distributed Collaborative Visual-inertial SLAM System for Aerial Swarm
293 stars 34 forks source link

Implementation for my custom dataset #45

Closed Eunseong17 closed 2 months ago

Eunseong17 commented 3 months ago

Thank you for the excellent work!

I am trying D2SLAM on my dataset, which consists of omnidirectional fisheye images.

The image topics in your dataset are composed of four images stitched together. I have the following questions:

  1. Are these composed image topics directly obtained from Arducam itself, or are they pre-stitched from each camera topic?

  2. If it is the latter, are the timestamps exactly identical? In my case, the cameras are synchronized via an internal trigger, but they are not perfectly identical, leading to different numbers of frames in the ROS bag file for each camera. I am curious about how you handled the timestamps during stitching.

Tis-Wu commented 2 months ago

HI~, You can refer to https://github.com/D2SLAM-Fusion/oak_ffc_4p_ros . I am trying to run D2SLAM using my own data, which consists of four fisheye cameras, and the final tracking results are terrible. I would like to ask how you performed the calibration?

Eunseong17 commented 2 months ago

Thank you very much.

I performed calibration using Tartancalib (https://github.com/castacks/tartancalib for camera intrinsic parameters) and Kalibr (for camera-IMU extrinsic parameters).