ICRA 2017 – Change Detection Datasets
This dataset suite presents three indoor change-detection scenarios collected by the Autonomous Systems Lab (ASL) at ETH Zurich. The corresponding publication is linked above.
Dataset Description
The dataset contains three indoor environments:
- office – A controlled office scene with four tripod static observations.
- lounge – A challenging meeting-area/office lounge, hand-held recordings over two weeks, frequent object changes, varying overlap ~50–100%.
- living_room – A baseline scene with nine hand-held recordings in a controlled indoor environment, high overlap and multiple viewpoints.
Each scenario provides sequences of RGB-D (and potentially point cloud) data, dense reconstructions, and change annotations for use in benchmarking change-detection and dynamic-object-discovery algorithms.
Data Access
- ETH Research Collection Link
- GitHub mirror / supplementary code: ethz-asl/change_detection_ds
Use Cases & Applications
- Long-term dense reconstruction and dynamic object discovery in indoor settings
- Benchmarking algorithms for change detection, TSDF-based scene modelling, dynamic scene segmentation
- Comparison of static vs. dynamic-object behaviour in indoor robotics environments
Citation
M. Fehr, F. Furrer, I. Dryanovski, J. Sturm, I. Gilitschenski, R. Siegwart, C. Cadena, “TSDF-based change detection for consistent long-term dense reconstruction and dynamic object discovery.”, IEEE International Conference on Robotics and Automation (ICRA) 2017, pp. 5237-5244. DOI: 10.3929/ethz-b-000721640
Tags: indoor, change-detection, TSDF, dynamic-objects