These steps were done in Ubuntu Server 16.04
SSH into AWS and do:
sudo vi /etc/ssh/sshd_config (or sudo nano /etc/ssh/sshd_config if you prefer nano)
Add line below to the file and then save it:
GatewayPorts clientspecified
Then run sudo service sshd restart
Additionally, open inbound ports 8080 and 5001 on AWS.
From the camera server, SSH into AWS EC2 again with
ssh -i YOUR_KEY.pem -R 0.0.0.0:8080:0.0.0.0:5000 ubuntu@YOUR_EC2_PUBLIC_IP
This will create a reverse tunnel so that connecting (inbound) on port 8080 on AWS will redirect to port 5000 on the camera server. This is equivalent to port forwarding port 5000 on the camera server network. Any IP will be able to access it.
This connection needs to be kept open the entire time, otherwise the tunneling will stop.
First install ffmpeg. This code relies on ffserver, which is no longer available in very recent versions of ffmpeg, so make sure you get something up to 3.4.x. I'm using ffmpeg 3.4.4.
I also had to install opencv using sudo apt-get install python-opencv (this will be different for a Mac) and pip install opencv-contrib-python
(just pip install opencv didn't work for me)
1 - On the camera server, run ffserver -f ffserver.conf. Leave this running.
2 - On another terminal tab, run ./start_ffmpeg.sh. Leave this running.
Note: I've only tested this on linux and this command might need changes to run on a Mac. I suspect you need to change -f v4l2 into -f avfoundation -i "default" in the start_ffmpeg.sh file.
If that doesn't work, try referring to the docs: https://trac.ffmpeg.org/wiki/Capture/Webcam
3 - On another terminal tab, run python app.py inside webm/flask_app_camera.
Now if you go to localhost:5000 you should see the stream.
Also, if you go to YOUR_EC2_PUBLIC_IP:8080, you should see the same stream again, this time going through AWS.
First you need to install required dependencies
$conda install ffmpeg=3.4 yes
$ffmpeg in your terminal
make sure it is 3.4 or 3.4.x and using conda ffmpeg
Open a new terminal
$cd webm/flask_app_camera folder
$ffserver -f ffserver.conf
Open a new terminal
$cd webm/flask_app_camera folder
$./start_ffmpeg.sh
Open a new terminal
$cd webm/flask_app_camera folder
$pip install svgwrite,if you don’t have it
$python app.py
Open a new tab in browser Go http://localhost:5000
- Clone the facenet repo https://github.com/davidsandberg/facenet
Also make sure to install their requirements https://github.com/davidsandberg/facenet/blob/master/requirements.txt
- Change the second line of each notebook in
facenet_tempto the path where you cloned the above repo + /src such as:
sys.path.append("/home/julia/USF/spring2/productAnalytics/facenet/src")
- Download pretrained model weights here https://drive.google.com/open?id=1R77HmFADxe87GmoLwzfgMu_HY0IhcyBz and unzip it so there's a folder named 20180408-102900 inside facenet_temp.
- Add everyones photos, make sure you have the below structure:
facenet_temp
├── 20180408-102900
│ ├── 20180408-102900.pb
│ ├── model-20180408-102900.ckpt-90.data-00000-of-00001
│ ├── model-20180408-102900.ckpt-90.index
│ └── model-20180408-102900.meta
├── Align images and create embeddings.ipynb
├── Facenet recognize frames.ipynb
└── photos
├── photos_raw
│ ├── byron
│ │ ├── UNADJUSTEDNONRAW_thumb_243f.png
...
- Run everything in
Align images and create embeddings.ipynb
This should create a pickle file containing everyone's embeddings
- Run everything in
Facenet recognize frames.ipynb
This should open your webcam and show your name when it recognized you
If you're testing the https://github.com/juliafeec/html5_video_streaming/tree/master/webm/flask_app_overlay app, you need to install Redis and start a Redis server.
On Mac:
- Install Redis:
$ pip install redis. - Start/initialize Redis server:
$ redis-server
To open a connection to Redis, append a key value pair to Redis storage and access this value from its key:
import redis
redis_db = redis.StrictRedis(host="localhost", port=6379, db=0)
redis_db.set('key', value)
redis_db.get('key')