Change Detection Datasets

Paper:

@inproceedings{fehr2017changedetection,
  title={{TSDF}-based Change Detection for Consistent Long-Term Dense
          Reconstruction and Dynamic Object Discovery},
  author="Fehr, Marius and Furrer, Fadri and Dryanovski Ivan and Sturm, Jurgen and Gilitschenski, Igor and Siegwart, Roland and Cadena, Cesar",
  booktitle={2017 IEEE International Conference on Robotics and Automation (ICRA)},
  year={2017},
  organization={IEEE}
}

DOWNLOAD (version 1.0)

Dataset Contents:

• For every observation:
  • Complete mesh of the aligned TSDF reconstructions.
  • Rosbag
    • T_G_C: Color camera transform (geometry_msgs/TransformStamped)
    • T_G_D: Depth camera transform (geometry_msgs/TransformStamped)
    • color_image: Color image (sensor_msgs/Image)
    • point_cloud_D: Point cloud in camera frame (sensor_msgs/PointCloud2)
    • point_cloud_G: Point cloud in global frame (sensor_msgs/PointCloud2)
    • tf: tf tree for world, depth_camera and color_camera (tf/tfMessage)

living_room

This is the baseline dataset and consists of 9 hand-held recordings in a controlled indoor environment. It provides nearly 100% observation overlap and also provides depth measurements from a large variety of viewpoints for most objects resulting in 3D models with a high coverage. The scene changes not only overlap in between observations but the dynamic objects also come in contact with different other objects.

office

This dataset consists of 4 observations of a controlled office environment recorded from a single point at the center of the room using a tripod. Hence the overlap of the observations is close to 100%. Its purpose is to be able to compare our approach to the meta-room algorithm which assumes a robotic platform with a pan-tilt RGB-D sensor unit that scans a single, convex room from a central point.

lounge

This is the most challenging dataset and consists of 10 hand-held recordings in an uncontrolled, challenging environment, a highly frequented meeting area/office lounge over the course of two weeks where objects are shifted on a daily basis. The observation overlap varies between approx. 50 - 100% and many dynamic objects are only partially observed.

Contact:

marius.fehr(at)mavt.ethz.ch