# Fall_Detection_Project

**Repository Path**: eden33s/Fall_Detection_Project

## Basic Information

- **Project Name**: Fall_Detection_Project
- **Description**: 此專案是針對一連串人體的骨幹座標(frames)變化來偵測人體的目前動作的 CNN，訓練資料來自 SBU_dataset 以及我自行利用拍攝好的照片經過 openpose 得出人體骨幹座標的資料
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2021-06-10
- **Last Updated**: 2021-06-10

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Fall Detection with Co-occurrence-Feature-Learning-From-Skeleton-2D-Data
***
此專案是針對一連串人體的骨幹座標(frames)變化來偵測人體的目前動作的 CNN，訓練資料來自 [SBU_dataset](https://www3.cs.stonybrook.edu/~kyun/research/kinect_interaction/index.html) 以及我自行利用拍攝好的照片經過 [openpose](https://github.com/CMU-Perceptual-Computing-Lab/openpose) 得出人體骨幹座標的資料
***
## Detail
-This is a project implemented with Keras based on the code from [here](https://github.com/fandulu/Keras-for-Co-occurrence-Feature-Learning-from-Skeleton-Data-for-Action-Recognition), [relative paper](https://arxiv.org/abs/1804.06055).

-Training and Testing data is in [data_seq dirctory](https://github.com/LinShien/Fall_Detection_Project/tree/master/data_seq).

-The whole model is defined in [model.py](https://github.com/LinShien/Fall_Detection_Project/blob/master/model.py).

-Addtional data-processing code is in [utils.py](https://github.com/LinShien/Fall_Detection_Project).

-To test the model, use [fall_detection.py](https://github.com/LinShien/Fall_Detection_Project/blob/master/fall_detection.py).

且部分資料由熱成像圖轉換成座標圖來做訓練，如下 :

class 9

![fall_img](./img/fall1.gif)
![fall_img2](./img/fall2.gif)

class 10

![sit_img](./img/sit1.gif)
![sit_img2](./img/sit2.gif)

## Result
Train with 416 skeleton sequences (20 frames)

Test with 83 skeleton sequences (20 frames)

Training acc : 100 %

Testing acc  : 94 %