https://github.com/chizhanyuefeng/Realtime-Fall-Detection-for-RNN
Real-time ADLs and Fall Detection implement TensorFlow
https://github.com/chizhanyuefeng/Realtime-Fall-Detection-for-RNN
adls fall fall-detection falldetector rnn tensorflow
Last synced: about 1 year ago
JSON representation
Real-time ADLs and Fall Detection implement TensorFlow
- Host: GitHub
- URL: https://github.com/chizhanyuefeng/Realtime-Fall-Detection-for-RNN
- Owner: chizhanyuefeng
- Created: 2018-06-15T02:32:19.000Z (almost 8 years ago)
- Default Branch: master
- Last Pushed: 2018-11-07T11:37:39.000Z (over 7 years ago)
- Last Synced: 2024-02-14T21:28:01.044Z (over 2 years ago)
- Topics: adls, fall, fall-detection, falldetector, rnn, tensorflow
- Language: Python
- Homepage:
- Size: 16.2 MB
- Stars: 89
- Watchers: 4
- Forks: 34
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Real-time Fall Detection for RNN(AFD-RNN)
result picture illustrate：
- The red,green,blue lines is acceleration sensor's x,y,z data。
- In the picture ,"correct" is the ground truth,"predict" is AFD-RNN network predict data
- Fall1、Fall2、Fall3 and Fall4 are represent Forward-lying,Front-knees-lying,Back-sitting-chair,Sideward-lying
## AFD-RNN using RNN
The sensors(acceleration and gyroscope sensor) is realtime to collect data,so we using rnn to detect the people movement.
## Requirenment
- TensorFlow >= 1.4
- python3
- matplotlib
## Class
Sitting,standing,stand to sit,sit to stand,upstairs,downstairs,lying,jumping,joging,walking and fall.
## Train and test
### 1.Train data
- The data collect frequence is 50Hz
- Need acceleration and gyroscope sensor
### 2.Before training
Put the train data to ./dataset/train/,and use kalman filter to handle the data.
python utils.py
### 3.Training
python train_rnn.py
## 4.Testing
Put the test data to ./dataset/test/,and use kalman filter to handle the data.
python run_rnn.py
## Dataset
We using public dataset [MobileFall](http://www.bmi.teicrete.gr/index.php/research/mobiact) to train and test our net.
I upload the dataset at [Baidu网盘](https://pan.baidu.com/s/1arZMNPs1GzWrQf4beJFCSQ),if you cant download from [MobileFall](http://www.bmi.teicrete.gr/index.php/research/mobiact),you can try this
The final accuracy is 98.78%