Fall Detection and Recognition 1: Fall Detection Dataset (with download link)
Table of contents
Fall Detection and Recognition 1: Fall Detection Dataset (with download link)
2. Fall posture: standing - bending over (squatting) - lying down
4. Falls Classification Dataset
5. Fall detection dataset download
6. Pytorch implements fall detection and recognition
7. Android implements fall detection and recognition
1 Introduction
This is the "Fall Detection Dataset (Including Download Link)" of the project " Fall Detection and Recognition " series of articles.
There are a lot of fall (wrestling, lying down) detection data on the Internet . I searched a lot in the CSDN download, but after downloading it, it was really unsatisfactory, and the quality was uneven. If you say too much, it will be tears, and it is a history of blood and tears. Lessons. A lot of data is very incomplete, either only labeling the fallen human body, or only labeling the standing human body, and the labeling format is not uniform.
In this article, we will share several data sets that can be used for fall detection and classification recognition; the Fall-Down-Detection-Dataset collects two: Fall-Down-Det-v1 and Fall-Down-Det -v2, a total of about 4000 pictures; Fall-Down-Classification-Dataset collected three: Fall-Down-Cls-v1, Fall-Down-Cls-v2 and Fall-Down-Cls -v3, 26000+ images in total. The data is of high quality and can be used for project model algorithm development for deep learning fall detection and recognition.
[Respect the principle, please indicate the source for reprinting] https://blog.csdn.net/guyuealian/article/details/130184256
For more articles on the "Fall Detection and Recognition" series, please refer to:
- Fall detection and recognition 1: Fall detection data set (including download link): https://blog.csdn.net/guyuealian/article/details/130184256
-
Fall detection and recognition 2: YOLOv5 realizes fall detection (including fall detection data set and training code): https://blog.csdn.net/guyuealian/article/details/130250738
-
Fall detection and recognition 3: Android implements fall detection (including source code, real-time fall detection): https://blog.csdn.net/guyuealian/article/details/130250824
-
Fall detection and recognition 4: C++ implementation of fall detection (including source code, real-time fall detection): https://blog.csdn.net/guyuealian/article/details/130250838
2. Fall posture: standing - bending over (squatting) - lying down
In the development of fall detection and recognition algorithms, we need to define the behavior state of falling. The project divides the fall state into three postures, namely: up (standing), bending (bending, squatting) and down (lying down) , In order to facilitate everyone's understanding, here are three graphic illustrations of postures
- up: such as human body standing, walking and other postures
- bending: postures such as bending over, squatting down, sitting down, etc.
- down: postures such as lying down, lying on the stomach, lying flat, etc.
3. Fall detection dataset:
The project has collected two fall detection datasets (Fall-Down-Detection-Dataset), Fall-Down-Det-v1 and Fall-Down-Det-v2, with a total of about 4000 images
（1）Fall-Down-Det-v1
Fall-Down-Det-v1 fall detection data set, the original pictures are mainly from surveillance videos, most of which are indoor rooms and other scenes, which are very in line with actual business scenarios.
A total of about 3400 pictures, divided into two subsets: training set (Train) and test set (Test); the total number of training sets (Train) exceeds 3253, and the total number of test sets (Test) is 210 Unified into the VOC data format, marking the three postures of the human body frame: up (standing), bending (bending, squatting) and down (lying down) , which can be directly used for deep learning object detection model training.
 |
 |
 |
 |
 |
 |
（2）Fall-Down-Det-v2
Fall-Down-Det-v2 fall detection data set, the original images are mainly from web crawler image retrieval, some of which are also integrated online data, the background is more complex, the number of people in the pictures is more, it can be used as fall detection data for general scenarios .
There are a total of 553 pictures, and the data annotation format is unified as the VOC data format. All the data have been marked with the human body detection frame, and the three postures of the human body frame are marked : up (standing), bending (bending, squatting) and down (lying down) Below) , which can be directly used for deep learning object detection model training.
 |
 |
 |
 |
4. Falls Classification Dataset
The project has collected three fall classification data sets (Fall-Down-Classification-Dataset), respectively Fall-Down-Cls-v1, Fall-Down-Cls-v2 and Fall-Down-Cls-v3, a total of 26000+ picture
（1）Fall-Down-Cls-v1
The Fall-Down-Cls-v1 fall classification data set is a fall classification data set made by cropping human body area pictures on the Fall-Down-Det-v1 fall detection data set. The cropped pictures are stored in up (standing), bending In the three folders (bending, squatting) and down (laying down), in order to enhance the generalization, each body frame area is respectively cropped according to the expansion of 1.0, 1.1 and 1.2 times ; divided into two sub- folders Collection: training set (Train) and test set (Test); the total number of training sets (Train) exceeds 9000+, and the total number of test sets (Test) is 600; all photos have been stored in their respective folders according to their categories, It can be directly used for deep learning classification model training.
| bending | down | up |
 |
 |
 |
（2）Fall-Down-Cls-v2
The Fall-Down-Cls-v2 fall classification data set is a fall classification data set made by cropping human body area pictures on the Fall-Down-Det-v2 fall detection data set. The cropped pictures are stored in up (standing), bending In the three folders (bending, squatting) and down (lying down), in order to enhance the generalization, each body frame area is cut out according to the expansion of 1.0, 1.1 and 1.2 times. Three pictures , a total of 6000+ pictures Pictures; all photos have been stored in their respective folders according to their categories, and can be directly used for deep learning classification model training.
| bending | down | up |
 |
 |
 |
（3）Fall-Down-Cls-v3
The Fall-Down-Cls-v3 fall classification data set, the original pictures are mainly from the image retrieval of web crawlers, the existing fall classification data on the Internet are integrated, and the data is cleaned. It is divided into two subsets: training set (Train) and test set (Test); the total number of training set (Train) exceeds 8000+, and the total number of test set (Test) is 700+; all photos have been stored in their respective categories Under the folder, it can be directly used for deep learning classification model training.
| bending | down | up |
 |
 |
 |
5. Fall detection dataset download
Fall dataset download address: Fall Detection and Recognition 1: Fall Detection Dataset (including download link)
The falls dataset contains:
- Fall detection data set: Fall-Down-Det-v1 and Fall-Down-Det-v2, a total of about 4000 images
- Fall classification dataset: Fall-Down-Cls-v1, Fall-Down-Cls-v2 and Fall-Down-Cls-v3, total 26000+ images
6. Pytorch implements fall detection and recognition
Reference article: Fall detection and recognition 2 "YOLOv5 realizes fall detection (including fall detection data set and training code)": https://blog.csdn.net/guyuealian/article/details/130250738
7. Android implements fall detection and recognition
Reference article: Fall detection and recognition 3: Android implements fall detection (including source code, real-time fall detection): https://blog.csdn.net/guyuealian/article/details/130250824
