This is an Undergraduate Final Year Project by Runze Cai. The goal of this FYP is to detect certain human behaviors based on OpenPose.
There are two main applications. The first one is Fall Detection, and the second is Detection of Bad Posture when Watching TV.
You may start to run the two applications with the following guide, and you can e-mail me cairunze17@mails.ucas.ac.cn if you meet any question.
Go to the folder ./model and run getModels.bat or getModels.sh （depend on your OS).
Go to this Website and download all the fall-**-cam0-rgb.zip files.
Put the zip files into the folder ./dataset/UR/ and unzip the images into ./dataset/UR/img.
For each different scene, you may select some images of the scene and put them into ./dataset/UR/bg to help the system
calculate the average background of this scene.
You may also download all the fall-**-cam0-rgb-d.zip files, and unzip the images into ./dataset/UR/depth.
Make sure your environment has OpenCV, Keras, Numpy, Pandas, Sklearn, Matplotlib, etc.
You can install all the requirements by enter pip install -r requirements.txt in your terminal.
Noted: be cautious about the path variables in files below.
To begin with, please run openpose_main.py to get Human Skeleton Points from the video frames. You can check the results
in the ./output/fall_output_mytest. You can get the results of rule-based fall detection as well.
At the same time, this python program generate .npy files which will be used later.
Run preprocess_points.py to preprocess the Human Skeleton Points for later training.
Run train.py to train the fall detection model based on skeleton points.
Run depth_train.py to train the fall detection model based on depth map.
Run merge_train.py to train the fall detection model with the combination of skeleton points and depth map.
Run fall_detection_demo.py to run the demo of fall detection.
The default model of fall detection in this program is model based on skeleton points. You may change the model if you like.
You may be required to input the information about e-mail of the sender and receiver when you run the program. And the sender will e-mail the receiver when the program detects fall.
Run watching_tv_pose_detection.py to run the demo of bad posture of watching TV detection.
You may use the video_to_img.py program to convert your video to image frames first. You may need to select an area which fit for the children watching TV first.
The Fall Detection program can detect whether the person stands or falls. And when the program detects fall, the program will send an e-mail the receiver.
The Detection of Bad Posture When Watching TV program can detect whether the person stay a normal posture or bad posture or sitting out of the selected area.
