Floor Detection

## Overview

Floor Detection is a dataset for object detection tasks - it contains Window Door annotations for 5,589 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Here are a few use cases for this project:

1. Retail Analytics: This model can be utilized to analyze the floor plan of retail stores. By identifying floors and shelves, it could provide data on product placement, customer's walking patterns, or best locations for advertising displays.

2. Robot Navigation: In warehouses or industrial settings, this model could be integrated into autonomous ground vehicles or robots to identify and navigate floor spaces, and avoid shelving units or other obstacles.

3. Layout Optimization: It can be used by architects, interior designers, or construction planners to optimize the use of space within a building. It can help identify whether a change in physical layout could improve functionality and usability.

4. Virtual Reality Space Modeling: Using this model, developers could create more realistic virtual environments for VR or AR applications such as simulations, games, or training programs.

5. Security and Surveillance: Integration into security systems to analyze human traffic, recognize unusual placements, or track an individual's behavior, which could be useful in both retail and security applications.

Floor Plan Object Detection

## Overview

Floor Plan Object Detection is a dataset for object detection tasks - it contains Floor Plan Objects annotations for 262 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Dirty Floor Detection

## Overview

Dirty Floor Detection is a dataset for instance segmentation tasks - it contains Muddy Floor annotations for 236 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Dataset for testing the performance of the number of floor detection model posted here.

Window Detection In Floor Plans

## Overview

Window Detection In Floor Plans is a dataset for object detection tasks - it contains Window annotations for 4,000 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

The global floor-standing laser intrusion detector market is experiencing robust growth, driven by increasing security concerns across various sectors, including military installations, port facilities, and critical infrastructure. The market, estimated at $150 million in 2025, is projected to witness a Compound Annual Growth Rate (CAGR) of 12% from 2025 to 2033. This expansion is fueled by technological advancements leading to enhanced detection accuracy and reliability, coupled with the rising adoption of sophisticated security systems in both public and private domains. Key trends include the miniaturization of devices for discreet deployment, the integration of advanced analytics for improved threat assessment, and the increasing demand for wireless connectivity for remote monitoring and management. Market segmentation reveals a strong preference for quad-beam detectors, followed by six-beam and eight-beam systems, reflecting the need to balance detection range and cost-effectiveness. The military sector represents a significant portion of the market, followed by platform and port applications, emphasizing the crucial role these detectors play in securing high-value assets. While the market faces restraints such as high initial investment costs and potential susceptibility to environmental factors like fog and dust, the overall growth trajectory remains positive, driven by continuous innovation and a growing awareness of security vulnerabilities. The major players in this market include Guangzhou AILIFU Electronic Technology co., Ltd., ATRD, Shandong Feitian Photoelectric Technology Co., Ltd, Shenzhen Alean Security Equipment Co., Ltd, VOANDOS, and 2M Technology Inc., amongst others. These companies are actively involved in research and development to enhance product features and expand market reach. Geographic analysis reveals strong growth across North America and Europe, driven by stringent security regulations and a high concentration of critical infrastructure. The Asia-Pacific region is expected to exhibit substantial growth potential due to rapid urbanization and increasing investments in security technologies. The market is anticipated to consolidate further in the coming years, with larger companies acquiring smaller players to gain a competitive edge and expand their product portfolio. This dynamic landscape promises continued innovation and growth within the floor-standing laser intrusion detector market.

Ground-level Blueberry Orchard Dataset v1 consists of 2000 RGB images of blueberry orchard scenes captured in the village of Babe, Serbia on three occasions in March, May, and August of 2022. Images are captured using the RGB module of Luxonis OAK-D device, with the resolution of 1920×1080 pixels and stored in the lossless PNG format.

The dataset is created for the purpose of training deep learning models for blueberry bush detection, for the task of autonomous UGV guidance. It contains sequences of images captured from the UGV moving and rotating in blueberry orchard rows. Images are captured from a height of approximately 0.5 meters, with the camera angled towards the base of a blueberry plant and the surrounding bank on which it grows. Dataset is captured in real-life outdoor conditions and contains multiple sources of variability (bush shape and size, lighting conditions, shadows, saturation etc.) and artifacts (occlusions by weeds, branches, presence of irregular objects etc.).

There are two classes of annotated objects of interest:

Bush, corresponding to the base of the blueberry bush.

Pole, corresponding to hail netting poles and similar obstructing objects such as lamp posts or wooden legs of bumblebee hives (distinguishing poles is important to prevent equipment damage in operations such as soil sampling and pruning).

Objects of interest are annotated with bounding boxes. Labels are saved in two formats:

LabelMe JSON format (x1, y1, x2, y2; in pixels)

Yolo TXT format (x_center, y_center, width, height; as a ratio of total image size, with numerical labels 0 and 1 corresponding to Bush and Pole)

There are 61 images with no annotated objects, and there are no corresponding label files for these images.

The dataset is split into train, validation and test sets with 75%, 10%, and 15% split (1490, 200, and 310 images, respectively). As the data contains sequences of images, the split is made based on sequences rather than individual images to prevent data leakage.

Detailed description and statistics are available in:

V. Filipović, D. Stefanović, N. Pajević, Ž. Grbović, N. Đurić and M. Panić, "Bush Detection for Vision-based UGV Guidance in Blueberry Orchards: Data Set and Methods," Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, Vancouver, Canada, 2023. (Accepted)

This is a dataset of ground truth annotations for benchmark data provided in A. Sielaff, D. Mangini, O. Kabov, M. Raza, A. Garivalis, M. Zupančič, S. Dehaeck, S. Evgenidis, C. Jacobs, D. Van Hoof, O. Oikonomidou, X. Zabulis, P. Karamaounas, A. Bender, F. Ronshin, M. Schinnerl,

J. Sebilleau, C. Colin, P. Di Marco, T. Karapantsios, I. Golobič, A. Rednikov, P. Colinet, P. Stephan, L. Tadrist, The multiscale boiling investigation on-board the international space station:

An overview, Applied Thermal Engineering 205 (2022) 117932. doi:10.1016/j.applthermaleng.2021.117932.

The annotations regard the 15 image sequences provided in the benchmark data and denoted as D1-D15.

The annotators were asked to localize the contact points and points on the bubble boundary so an adequate contour identification is provided, according to the judgement of the expert. The annotators were two multiphase dynamics experts (RO, SE) and one image processing expert (ICS). The annotators used custom-made software to pinpoint samples upon contour locations in the images carefully, using magnification, undo, and editing facilities. The experts annotated the contact points and multiple points on the contour of the bubble until they were satisfied with the result.

The annotations were collected for the first bubble of each sequence. For each bubble, 20 frames were sampled in chronological order and in equidistant temporal steps and annotated. All experts annotated data sets D1-D15. The rest were annotated by ICS after learning annotation insights from the multiphase dynamics experts.

The format of the dataset is as follows. A directory is dedicated to each bubble annotation. The directory name notes the number of the dataset and the annotator id. Each directory contains 20 text files and 20, corresponding, images. Each text file contains a list with the 2D coordinates of one bubble annotation. The first coordinate marks the left contact point and the last coordinate marks the right contact point. These coordinates refer to a corresponding image contained in the same directory. Text files and image files are corresponded through their file names, which contain the frame number. The frame number refers to the image sequence. Images are in lossless PNG format.

The global floor covering market, valued at $96.85 billion in 2025, is projected to experience robust growth, exhibiting a Compound Annual Growth Rate (CAGR) of 4.91% from 2025 to 2033. Several key drivers fuel this expansion. Increasing construction activities, particularly in residential and commercial sectors across rapidly developing economies in Asia-Pacific and other regions, significantly boost demand. Furthermore, rising disposable incomes, coupled with a growing preference for aesthetically pleasing and durable flooring solutions, contribute to market growth. Significant trends shaping the market include the surge in popularity of sustainable and eco-friendly flooring options, such as recycled materials and low-VOC products, driven by growing environmental awareness. Technological advancements, like the introduction of innovative materials and improved manufacturing processes, are also contributing to product diversification and enhanced performance characteristics. However, the market faces certain restraints, including fluctuating raw material prices (e.g., petroleum-based polymers) and economic downturns that can impact construction activity. The market is segmented geographically, with North America, Europe, and Asia-Pacific representing significant regional markets. Analysis across various segments—production, consumption, import/export volumes and values, and price trends—reveals regional disparities in manufacturing capabilities, consumption patterns, and trade dynamics. Key players like Shaw Industries Group Inc., Mohawk Industries Inc., Tarkett Group, and Armstrong Flooring compete through product innovation, brand recognition, and strategic acquisitions, while regional players cater to local market needs and preferences. The market's segmentation extends beyond geographical boundaries to include various product types (carpet, hardwood, vinyl, ceramic tile, etc.), each with its own growth trajectory and market dynamics. Detailed analysis of these segments within each region provides a comprehensive understanding of supply-demand relationships and price fluctuations. For instance, the demand for resilient flooring (vinyl, linoleum) is increasing due to its affordability and durability, while the hardwood flooring segment benefits from its aesthetic appeal and perceived higher value. Import and export data highlight the global nature of the floor covering industry, with certain regions specializing in manufacturing specific types of flooring and others acting as major importers to satisfy domestic demand. The price trend analysis is crucial in understanding the impact of raw material costs, manufacturing efficiencies, and competitive pressures on market profitability. Companies are continuously striving to enhance their product portfolios, focusing on customization options and improving their supply chain to navigate challenges related to raw material availability and transportation costs. This analysis, incorporating data from 2019-2024, provides a strong foundation for forecasting market trends through 2033, offering valuable insights for investors, manufacturers, and stakeholders across the global floor covering industry. Recent developments include: June 2023: The Icon Group, a construction company that specializes in flooring and lockers for athletic facilities and surfaces for equestrian and other animal purposes, was formed by the merger of Abacus Sports Installations, Rubber Flooring Systems in Kemah, and Spec Athletic, which were already providers of sports flooring., April 2023: BIG signed with the Australian B2B flooring distributor to acquire Signature Floors' broad range of operations. This acquisition increased their prospects for expansion in the soft, resilient, and hard flooring markets in Australia and New Zealand.. Key drivers for this market are: Rising Residential Space Driving The Market, Rising Urbanization Driving Demand for Floor Coverings. Potential restraints include: Impact on Real Estate Construction Market Affecting Floor Covering Demand, Volatility in Price of Raw Material Affecting the Market. Notable trends are: Rising Sales Of Different Types Of Floor Covering.

The purpose of this dataset is to evaluate the performance of floor vibration sensing in tracking the progression of muscular dystrophy through individuals’ gait patterns. We recruited human subjects (N=36) and conducted experiments at Stanford University and Nationwide Children's Hospital in Columbus, Ohio, with healthy human subjects (N=21) and children with MD (N=15).

The experiments consist of two phases - Phase 1 is a pilot study to test the feasibility of floor vibration sensing in capturing gait characteristics in lab settings; Phase 2 has two hospital studies to evaluate our sensing method in real life. This dataset consists of floor vibration data (vertical velocity of floor vibration) induced by 9 healthy subjects’ footsteps from these two phases, named lab data.zip, hospital data 1.zip, and hospital data 2.zip, respectively. The complete dataset requires a data-sharing agreement with Nationwide Children's Hospital.

The custom coding scripts (in MATLAB and Python) to process the data are included in the code.zip file, which includes 3 processing steps: 1) Preprocessing and Footstep Detection, 2) Feature Extraction, and 3) Model Prediction, the scripts of each step are grouped into a folder with their corresponding names. Detailed function descriptions are included in the scripts.

The footstep-induced structural vibration data is stored as both raw data and as individual footsteps. The raw data files start with “raw_”, each consisting of a series of consecutive footsteps (see the sample plot). The individual footstep data files start with “detected_steps_”, each consisting of one single footstep detected from the raw data. The dataset is stored in MAT file format that can be accessed through MATLAB.

The sensing unit consists of 5 components: 1) the geophone (SM-24), 2) the amplification module, 3) the processor board, 4) the data acquisition module (NI-Daq), and 5) the power cables. The sensing unit converts the structural vibration velocity into voltage records. The sampling frequency is 500 Hz for the lab study and 25600 Hz for the hospital studies.

The experiment setup, sample data plot, and code usage can be found in Dataset Description.pdf. For more details about the hospital studies, please refer to the MD-Vibe paper at the following link: https://doi.org/10.1145/3410530.3414610

Please cite this dataset as:

Yiwen Dong, Megan Iammarino, Jingxiao Liu, Jesse Codling, Jonathon Fagert, Mostafa Mirshekari, Linda Lowes, Pei Zhang, and Hae Young Noh. 2023. The MD-Vibe Dataset: Footstep-Induced Floor Vibration Data for Functional Gait Assessment in Individuals with Muscular Dystrophy. Zenodo, DOI: https://doi.org/10.5281/zenodo.8125704

Yiwen Dong, Joanna Jiaqi Zou, Jingxiao Liu, Jonathon Fagert, Mostafa Mirshekari, Linda Lowes, Megan Iammarino, Pei Zhang, and Hae Young Noh. 2020. MD-Vibe: physics-informed analysis of patient-induced structural vibration data for monitoring gait health in individuals with muscular dystrophy. In Adjunct Proceedings of the 2020 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2020 ACM International Symposium on Wearable Computers (UbiComp-ISWC '20). Association for Computing Machinery, New York, NY, USA, 525–531. https://doi.org/10.1145/3410530.3414610

Overview

The Walker Fall Detection Data Set is a curated compilation of inertial data designed for the study of fall detection systems, specifically for people using walking assistance. This data set offers deep insight into various movement patterns. It covers data from four different classes: idle, motion, step and fall.

This dataset was published as part of a research paper: Dataset and System Design for Orthopedic Walker Fall Detection and Activity Logging Using Motion Classification

Data Acquisition

Data was recorded using an IMU affixed to a walker, as illustrated in the image below:

https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F15943143%2F134aaf57e284d20775d37ecdfa14d30e%2F20230712_163957.jpg?generation=1695049359163422&alt=media" alt="Walker">

The IMU used for this project is the Arduino Nano 33 BLE Sense. It's powered by a LiPo battery and is equipped with a voltage regulator and a dedicated battery charging circuit. To ensure durability and protection during the data recording phase, the entire prototype was securely housed in a custom 3D-printed casing.

https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F15943143%2F0c8458d7bd161e983e1676e72d89085d%2F20230712_163957_2.jpg?generation=1695050869941377&alt=media" alt="">

The prototype was designed to transmit data wirelessly to a computer using Bluetooth Low Energy (BLE). Upon receipt, a Python script processed the incoming data and stored it in JSON format. The data transmission rate was optimized to achieve the highest possible rate, resulting in approximately 100 samples per second, covering both accelerometer and gyroscope data.

Data were collected from four different subjects, each of whom maneuvered the walker down a hallway, primarily capturing step and movement data. It is important to note that the “**idle**” data are not subject-specific, as it represents periods in which the walker is stationary. Similarly, “**fall**” data is also not linked to any particular individual; was obtained by deliberately pushing the walker from a vertical position to the ground.

Data Processing

This dataset contains four classes:

• Idle: Represents periods of no movement, indicating that the walker is stationary.
• Step: Capture moments when an individual takes a step using the walker.
• Movement: Covers any movement that is not considered a step or a fall, such as when carrying the walker or adjusting its position.
• Fall: Denotes cases in which the walker tips over and falls to the ground from an upright position.

To effectively categorize the data, several processing steps were executed. Initially, the data was reduced from its original 100 samples per second to ensure a constant time step between samples, since the original rate was not uniformly constant. After this, both the acceleration and gyro data were normalized to one sample every 12.5 milliseconds, resulting in a rate of 80 samples per second. This normalization allowed the synchronization of acceleration and gyroscope data, which were subsequently stored in dictionaries in JSON format. Each dictionary contains the six dimensions (three acceleration and three gyro) corresponding to a specific timestamp.

To distinguish individual samples within each group, the root mean square (RMS) value of the six dimensions (comprising acceleration and gyroscope data) was calculated. Subsequently, an algorithm based on the hidden Markov model (HMM) was used to discern the hidden states inherent in the data, which facilitated the segmentation of the data set.

https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F15943143%2F250182bc7b0c3ca0383ea79ac6e3224a%2Fhmm.jpg?generation=1695059033833835&alt=media" alt="">

Through the filtering process, the HMM effectively identifies individual steps. Once all steps were identified, the window size was determined based on the duration of each step. A window size of 160 samples was chosen, which, given a rate of 80 samples per second, is equivalent to a duration of 2 seconds for each sample.

A similar procedure is employed to extract "fall" samples. However, for "idle" and "motion" samples, isolation isn't necessary. Instead, samples from these categories can be arbitrarily chosen from the recorded clusters.

Final Dataset The finalized dataset is presented in CSV format. The first column serves as the label column and covers all four classes. In addition to this, the CSV file has 960 columns of functions. These columns encapsulate 160 samples each of acceleration and gyro data in the x, y, and z axes.

Each class contains 620 samples, bringing the overall total to 2480 samples across all classes.

Citation and Use

This dataset is associated with a research article currently un...

d1—d2 represents the conformity among the different goodness metrics (purity, NMI and ARI) in terms of Kendall’s and Spearman’s rank correlation (see text). The last column reports the Kendall’s τ and Spearman’s ρ rank correlations of with the majority ranking of similarity to the ground truth (see text).

Floor Plan Connection Detection

## Overview

Floor Plan Connection Detection is a dataset for object detection tasks - it contains Door annotations for 395 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

The investigation of indoor navigation system presents an intriguing study especially in the field of floor cleaning robot research. Experiments involving wall detection and motion control test demonstrate that the robot can be implemented for domestic purpose. This experiment is conducted for verifying the performance of ultrasonic sensor. Robot is placed near the wall and the sensor value output is observed. The robot is tested in ten points along the wall. The test is performed to evaluate sensor accuracy while maintaining close proximity to the wall. This hallway provides a challenging task for the robot since the distance between robot and wall is not constant. While the hallway wide is about 2 meter, the doors and building concrete column are expected to provide various sensor measurement result. The result shows relative linear association of actual distance with sensor value. First column of the table conveys sensor 1 distance measurement data from ten points along hallway. The average measured distance value is 38.928 cm. While the real distance value between robot and wall is 40 cm, the average distance error is 1.072 cm. Another result from the observation was that sensor 2, 3 and 4 measurement data has provided slightly better accuracy than other sensors with error margin 0.074, 0.01, and 0.302 respectively.

Performance measurements of object detections for YOLOv5 VS YOLOv8.

This image dataset: "Sentinel-1 SAR Oil spill image dataset for train, validate, and test deep learning models. Part III", is the third part of the image dataset for train and validate deep learning models for oil spill detection and segmentation.

This part contains the test images.

The dataset comprises Sentinel-1 SAR images in Sigma0, in decibels (db), along with their ground truth. The images are 2048x2048x2, also the ground truth is 2048x2048; all of them are in TIFF format.

The files are organized in the following manner:

• Images: Sentinel-1 SAR images in Sigma0 in decibels.
  • Lookalike: There are 150 images corresponding to look-alike surfaces
  • No oil: There are 150 oil free images
  • Oil: There are 150 oil spill images
• Mask: Gorund truth masks images for each Sentinel-1 SAR images
  • Lookalike: There are 150 images corresponding to the ground truth for look-alike images
  • No oil: There are 150 images to corresponds oil free ground truth images
  • Oil: There are 150 images for the ground truth for oil spill images

Each corresponding ground truth has the same number as its respective image. For instance, the image of an oil spill has a corresponding number of 0001, as well as its ground truth.

The complete dataset consists of three parts:

Sentinel-1 SAR Oil spill image dataset for train, validate, and test deep learning models. Part I. (10.5281/zenodo.8346860)

Sentinel-1 SAR Oil spill image dataset for train, validate, and test deep learning models. Part II. (10.5281/zenodo.8253899)

Sentinel-1 SAR Oil spill image dataset for train, validate, and test deep learning models. Part III. (10.5281/zenodo.13761290)

The global synthetic sports floor system market is experiencing robust growth, driven by increasing participation in sports and fitness activities, coupled with the rising preference for durable, high-performance flooring solutions in various settings. The market's expansion is fueled by several key factors, including the construction of new sports arenas, schools, and fitness centers globally, as well as renovations and upgrades in existing facilities. Technological advancements leading to improved shock absorption, enhanced durability, and aesthetically pleasing designs further stimulate market demand. The polymeric floor segment currently holds a significant market share due to its versatility and cost-effectiveness, while the rubber floor segment is witnessing strong growth driven by its superior impact absorption properties, making it ideal for high-impact activities. Key regional markets include North America and Europe, driven by high disposable incomes and robust infrastructure development. However, factors such as high initial investment costs for installation and potential maintenance requirements can act as restraints to market growth. The market is expected to witness increased competition among key players, such as Tarkett, Armstrong, and Mondo, focusing on product innovation, strategic partnerships, and expansion into new geographic markets. Over the next decade, the market is poised for significant growth, driven by factors such as government initiatives promoting sports infrastructure development and the growing adoption of synthetic sports flooring in developing economies. The competitive landscape is characterized by a mix of established global players and regional manufacturers. The established players possess strong brand recognition and extensive distribution networks, while regional players cater to specific geographic needs and offer competitive pricing. The market is witnessing increased collaboration between flooring manufacturers and sports facility designers to offer integrated solutions, incorporating technological advancements in flooring materials and installation techniques. Future growth will be shaped by technological innovations like the development of sustainable and environmentally friendly materials, and increased focus on customized flooring solutions tailored to specific sports and user needs. The market will also see continued expansion into emerging markets in Asia-Pacific and Latin America, fueled by rapid urbanization, economic growth, and increased investment in sports infrastructure. Overall, the synthetic sports floor system market presents a promising investment opportunity for players who can effectively adapt to evolving customer needs and technological advancements.

The global automatic floor scraping machine market is experiencing robust growth, driven by increasing demand across commercial and residential sectors. The rising adoption of these machines in various applications, including large-scale industrial facilities, supermarkets, and even residential homes for improved floor maintenance, fuels this expansion. Technological advancements, such as improved efficiency, reduced noise levels, and enhanced safety features, are further propelling market adoption. While the exact market size in 2025 requires further specification, considering a plausible CAGR of 5% based on industry trends and a starting point of $500 million in 2019 (an assumed figure based on common market sizes for niche equipment), we can project a 2025 market size of roughly $700 million. This growth is expected to continue throughout the forecast period (2025-2033), with steady expansion across different segments. The walk-behind segment currently holds a significant market share due to its cost-effectiveness and suitability for smaller spaces, but ride-on and self-propelled machines are gaining traction owing to their higher efficiency in large areas. Geographical distribution shows strong growth in North America and Europe, fueled by high construction activity and stringent building regulations promoting clean and well-maintained facilities. However, increasing competition from emerging manufacturers and potential economic fluctuations represent key restraints to consider. The market segmentation reveals promising opportunities for specialized manufacturers. For instance, focusing on the growing residential segment by offering compact and user-friendly models could be a lucrative strategy. Similarly, innovation in sustainable and energy-efficient models will attract environmentally conscious buyers. The competitive landscape is characterized by a mix of established players and emerging companies. While established brands benefit from brand recognition and a strong distribution network, innovative startups are challenging the market with cost-effective solutions and cutting-edge technologies. A strategic approach involving product differentiation, technological innovation, and effective marketing campaigns will be crucial for success in this dynamic market. Furthermore, focusing on expansion into developing economies with growing infrastructure projects offers significant potential for future growth.

This dataset that can be used to evaluate methods, which are able to detect changed objects when comparing two recordings of the same environment at different time instances. Based on the labeled ground truth objects, it is possible to differentiate between static, moved, removed and novel objects.

Dataset Description

The dataset was recorded with an Asus Xtion PRO Live mounted on the HSR robot. We provide scenes from five different rooms or parts of rooms, namely a big room, a small room, a living area, a kitchen counter and an office desk. Each room is visited by the robot at least five times while between each run a subset of objects from the YCB Object and Model Set (YCB)[1] is re-arranged in the room. In total we generated 26 recordings. For each recording between three and 17 objects are placed (219 in total). Furthermore, furniture and permanent background objects are slightly rearranged. These changes are not labeled because for most service robot tasks, this is not relevant.

Assuming most objects are placed on horizontal surfaces, we extracted planes in each room in a pre-processing step (excluding the floor). For each surface, all frames from the recording where it is visible are extracted and used as the input for ElasticFusion[2]. This results in a total of 34 reconstructed surfaces.

We provide pointwise annotation of the YCB objects for each surface reconstruction from each recording.

Images of exemplary surface reconstructions can be found here: https://www.acin.tuwien.ac.at/vision-for-robotics/software-tools/obchange/

Dataset Structure

The file structure of ObChange.zip is the following:

Room

- scene2

- planes

- 0

- merged_plane_clouds_ds002.pcd

- merged_plane_clouds_ds002.anno

- merged_plane_clouds_ds002_GT.anno

- 1

- merged_plane_clouds_ds002.pcd

- merged_plane_clouds_ds002.anno

- merged_plane_clouds_ds002_GT.anno

- ...

table.txt

- scene3

The pcd-file contains the reconstruction of the surface. The merged_plane_clouds_ds002.anno lists the YCB objects visible in the reconstruction and merged_plane_clouds_ds002_GT.anno contains the point indices of the reconstruction corresponding to the YCB objects together with the corresponding object name. The last element for each object is a bool value indicating if the object is on the floor (and was reconstructed by chance). The table.txt lists for each detected plane the centroid, height, convex hull points and plane coefficients.

We provide the original input data for each room. The zip-files contain the rosbag file for each recording. Each rosbag contains the tf-tree and the RGB and depth stream, as well as the camera intrinsic. Additionally, the semantically annotated Voxblox[3] reconstruction created with SparseConvNet[4] is provided for each recording.

You may also be interested in Object Change Detection Dataset of Indoor Environments. It uses the same input data, but the ground truth annotation is based on a full room reconstruction instead of individual planes.

Acknowledgements

The research leading to these results has received funding from the Austrian Science Fund (FWF) under grant agreement Nos. I3969-N30 (InDex), I3967-N30 (BURG) and from the Austrian Research Promotion Agency (FFG) under grant agreement 879878 (K4R).

References

[1] B. Calli, A. Singh, J. Bruce, A. Walsman, K. Konolige, S. Srinivasa, P. Abbeel, A. M. Dollar, Yale-CMU-Berkeley dataset for robotic manipulation research, The International Journal of Robotics Research, vol. 36, Issue 3, pp. 261 – 268, April 2017.

[2] T. Whelan, S. Leutenegger, R. Salas-Moreno, B. Glocker, A. Davison, ElasticFusion: Dense SLAM without a pose graph, Proceedings of Robotics: Science and Systems, July 2015.

[3] H. Oleynikova, Z. Taylor, M. Fehr, R. Siegwart, J. Nieto, Juan, Voxblox: Incremental 3D Euclidean Signed Distance Fields for On-Board MAV Planning, in Proceedings of IEEE International Conference on Intelligent Robots and Systems (IROS), pp. 1366-1373, 2017.

[4] B. Graham, M. Engelcke, L. van der Maaten, 3D Semantic Segmentation with Submanifold Sparse Convolutional Networks, Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), pp. 9224 – 9232, 2018.