Indoor Dataset for Place Categorization

RGB-D Sensor

Description

This data consists of RGB-D images taken by a Kinect Camera on a rotating base at a height of 125 cm. The Kinect sensor has 43 degeees vertical by 57 degrees horizontal field of view, and a depth sensor range of 1.2m - 3.5m.

            



Data format

Each depth image is represented by a 480 x 640 matrix containing depth data in a .txt file, with undefined values represented by NaN. In addition, the corresponding grayscale image is provided where NaN values are represented by full black.

Each corresponding RGB image stored in PNG format.



Place Categories

The RGB-D images were obtained at 6 different place categories in different buildings at Fukuoka, Japan. Categories include corridor, kitchen, office, study room, and toilet. Each category contains different sets of RGB-D images. Each set was obtained in a different place inside the same category.



Dataset

Data is stored using the following directory and files structure, where names in [brackets] indicate variables:

	
		[Category]  # Directory with the category Category
			|
			----- [Building]_[FloorId]_[Category]_[PlaceId]  # Directory containing RGB-D images
					|
					|
					------- depth_image[ImageId].txt # Depth information in TXT format.
					|
					|
					------- depth_image[ImageId].png # Depth information in PNG format.
					|
					|
					------- image[ImageId].png       # RGB data in PNG format


	

The following table shows the distribution of the data set. An example is provided for each place category.

RGB-D Indoor Datasets
Category
Example Color Image (click to enlarge)
corridors
(255 Mbyte,
5 places)
corridor (example)
(655kB)
kitchens
(204 Mbyte,
8 places)
kitchen (example)
(519kB)
labs
(583 Mbyte,
4 places)
labs (example)
(550kB)
offices
(95 Mbyte,
3 places)
offices (example)
(621kB)
studyrooms
(328 Mbyte,
8 categories)
study rooms (example)
(621kB)
toilets
(116 Mbyte, 3 places)
toilets (example)
(621kB)