Overview:
The datasets that are used for the simulation purpose are
raw RGB and Depth images of size 320x240 recorded from a single uncalibrated
Kinect sensor after resizing from 640x480. The Kinect sensor is fixed at roof
height of approx 2.4m. The datasets contain a total of
21499 images. Out of total datasets of 22636 images, 16794 images can be used for training,
3299 images can be used for validation and
2543 images can be used for the test. The images in the dataset are
recorded in 5 different rooms which consist of 8 different view angles. There
are 5 different participants out of which there are two male participants of age
32 and 50 and three female participants of age 19, 28 and 40. All the activities
of the participants represent 5 different categories of poses that are standing,
sitting, lying, bending and crawling. There is only one participant in each
image. Some images in the datasets are empty which are categorised as 'other'.
We have used images of 2 participants: the male of age 32 and the female of age
28 combining total of 16794 images for training, and 3299 images for validation
which contains a male participant of age 32 from training set but is in a
different room to that of training and testing set. Similarly, the test set
contains images of 3 participants out of which 2 female participants are of age
19 and 40 and a male participant is of age 50. These images are recorded in a
different room that is not seen in training or validation set. These total of
22636 images are in sequence but have not repeated anywhere in the sequence and
all the sets have original and its horizontal flipped images added in sequence
to increase the number of images in a set.
Train Dataset(RGB+Depth+Label) |
Validate Dataset(RGB+Depth+Label) |
Test Dataset(RGB+Depth+Label) |
|
||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Total:16794 |
Total:3299 |
Total:2543 |
S
Serial number | Class |
1 | 4 |
2 | 3 |
3 | 0 |
4 | 2 |
5 | 1 |
6 | 5 |
This dataset have been created for a research work that
aims for a computer vision based indoor fall detection. The idea to create the
dataset is to recognise specific poses that helps in understanding the fall
incident. We would like to share this data which could be useful for other
academic purposes in the future.