├── README.md └── previews ├── living_room.png ├── lounge.png └── office.png /README.md: -------------------------------------------------------------------------------- 1 | # Change Detection Datasets 2 | 3 | Paper: 4 | ``` 5 | @inproceedings{fehr2017changedetection, 6 | title={{TSDF}-based Change Detection for Consistent Long-Term Dense 7 | Reconstruction and Dynamic Object Discovery}, 8 | author="Fehr, Marius and Furrer, Fadri and Dryanovski Ivan and Sturm, Jurgen and Gilitschenski, Igor and Siegwart, Roland and Cadena, Cesar", 9 | booktitle={2017 IEEE International Conference on Robotics and Automation (ICRA)}, 10 | year={2017}, 11 | organization={IEEE} 12 | } 13 | ``` 14 | 15 | ## [DOWNLOAD (version 1.0)](http://robotics.ethz.ch/~asl-datasets/icra_2017_change_detection/) 16 | 17 | **Dataset Contents:** 18 | * For every observation: 19 | * Complete mesh of the aligned TSDF reconstructions. 20 | * Rosbag 21 | * `T_G_C`: Color camera transform (*geometry_msgs/TransformStamped*) 22 | * `T_G_D`: Depth camera transform (*geometry_msgs/TransformStamped*) 23 | * `color_image`: Color image (*sensor_msgs/Image*) 24 | * `point_cloud_D`: Point cloud in camera frame (*sensor_msgs/PointCloud2*) 25 | * `point_cloud_G`: Point cloud in global frame (*sensor_msgs/PointCloud2*) 26 | * `tf`: tf tree for `world`, `depth_camera` and `color_camera` (*tf/tfMessage*) 27 | 28 | 29 | 30 | ## living_room 31 | 32 | ![living_room](https://github.com/ethz-asl/change_detection_ds/blob/master/previews/living_room.png?raw=true) 33 | 34 | 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. 35 | 36 | ## office 37 | 38 | ![office](https://raw.githubusercontent.com/ethz-asl/change_detection_ds/a854383e2ba7d706db46876ce5fe93b5bf50a7ad/previews/office.png) 39 | 40 | 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. 41 | 42 | ## lounge 43 | 44 | ![lounge](https://github.com/ethz-asl/change_detection_ds/blob/master/previews/lounge.png?raw=true) 45 | 46 | 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. 47 | 48 | ## Contact: 49 | marius.fehr(at)mavt.ethz.ch 50 | -------------------------------------------------------------------------------- /previews/living_room.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/ethz-asl/change_detection_ds/592742ceeeed8cf102fd1a7eb4f01be093516ec0/previews/living_room.png -------------------------------------------------------------------------------- /previews/lounge.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/ethz-asl/change_detection_ds/592742ceeeed8cf102fd1a7eb4f01be093516ec0/previews/lounge.png -------------------------------------------------------------------------------- /previews/office.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/ethz-asl/change_detection_ds/592742ceeeed8cf102fd1a7eb4f01be093516ec0/previews/office.png --------------------------------------------------------------------------------