Notice how I said live cam and not live stream? It's because we'll be dealing with normal images. Go back about 15 years and this technology was exciting. Live weather cams and nature cams were deadly cool. Image-based live streams are also a little more robust when the connection between our Raspberry Pi and web server can be unreliable (e.g., WiFi).
This method can hit around 5-10fps depending on the image resolution. There's a
benchcamera.py script in the repository that you can run on your Pi to test. Full installation instructions for this project can also be found in the repo.
Let's work back to front, starting with the web server. The Flask app we'll be coding can be hosted anywhere. Its job is to receive images from the Pi, store the latest image and serve this image to clients. Temporarily storing this image is the trickiest part of this project. We want our app to be scalable to, say, a few hundred clients. Flask apps are normally hosted via servers like uWSGI or Gunicorn and they handle the load in part by using threads.
A different thread of the app is created for each user. These instances can't share state at the application level. We solve this by storing the image data in a local SQLite database called
global.db. Dealing with multiple readers and writers is easy for databases. They handle it by using locks. This ensures that our Flask app won't ask for an image that is halfway through being written to the database and then send corrupted data to the client.
# app.pyimport osimport sqlite3from flask import Flask, request, gapp = Flask(__name__)DATABASE = 'global.db'# helper method, allows database access within a controllerdef get_db():db = getattr(g, '_database', None)if db is None:db = g._database = sqlite3.connect(DATABASE)return db# create our database structure, which is akin to a dict with one keydef init_db():db = sqlite3.connect(DATABASE)cur = db.cursor()cur.execute('''CREATE TABLE IF NOT EXISTS store(id INTEGER PRIMARY KEY, image BLOB)''')cur.execute("INSERT OR IGNORE INTO store (id, image) VALUES (1, '')")db.commit()db.close()# initialize dbinit_db()# close the database connection after every request firstname.lastname@example.org_appcontextdef close_connection(exception):db = getattr(g, '_database', None)if db is not None:db.close()
When we start our app, we need to make sure that there is a table, and a slot in that table to store images. At the same time, we don't want to overwrite existing images. So we create a table only if it doesn't exist already
CREATE TABLE IF NOT EXISTS and then we create the one row that we will be using with similar logic
INSERT OR IGNORE.
The upload controller (also known as a handler) exists on the root path and only accepts POST requests. It also checks for an environmental value called
PASSWORD. (Query parameters are insecure over HTTP but more secure over HTTPS).
# recieve images and write to db as BLOB if the password is email@example.com('/', methods=['POST'])def update_image():db = get_db()cur = db.cursor()if request.args.get('password') != os.environ['PASSWORD']:return '', 400else:image = [request.data]cur.execute("UPDATE store SET image=? WHERE id=1", image)db.commit()return '', 200
We can now receive images! We will only store one image: the latest one. Serving that image to any client that makes a request is straightforward.
# share images naively and let browsers interpret the BLOB as firstname.lastname@example.org('/live.jpeg')def get_image():cur = get_db().cursor()image = cur.execute("SELECT image FROM store WHERE id=1").fetchone()return image, 200
If we host a link to this image as
setInterval. We can use this logic to build a live cam widget. Our test route uses this so we can make sure our images are being received.
# test route that mimics having a live email@example.com('/test')def test_image():return '''<img src="/live.jpeg" /><script>setInterval(() =>document.querySelector(\'img\').src = \'/live.jpeg?\' + Date.now(), 150)</script>'''
We cache-bust to make sure that the browser doesn't think that it already has the image source in its cache. One way to cache-bust is to add a unique query string to the end of a request (that ultimately does nothing) like the current date in milliseconds:
$ PASSWORD='123'$ FLASK_APP=app.py flask run* Running on http://localhost:5000/
For installation and running questions related to Flask, see their docs.
Raspberry Pi Script
The script that runs on the Pi will capture images and send them via a POST request. We install our modules with
pip install requests picamera. The script is run via
python camera.py or
# camera.pyimport ioimport timeimport requestsimport picamera# change this line to your Flask app's address!url = 'http://192.168.1.100:5000/?password=123'framerate = 90quality = 100res = (1280, 720)with picamera.PiCamera(framerate=framerate, resolution=res) as camera:time.sleep(2) # camera warm-up timewhile True:try:image = io.BytesIO()camera.capture(image, 'jpeg', quality=quality, use_video_port=True)r = requests.post(url, data=image.getvalue())# place a `time.sleep` here if you want a slower live camexcept:time.sleep(5) # wait for WiFi/server to come back
You will probably want to run this as a startup script for maximum uptime. If there's a network problem, the script will keep attempting to post an image until it gets through. If there's a problem with your Pi, our Flask app will keep serving the latest image it received. Robust!
Some thoughts to improve the project but make it a worse tutorial.
- Add a listener to SQLite and cache the latest image in memory.
- Upload images via WebSocket (save bandwidth and latency).
- Serve images to clients over WebSocket as they arrive (same as above).
- Use Flask for uploading images, serve images as a static file via HTTP server (scales infinitely).
- When receiving images, scale them to different resolutions (e.g.,
Reach out, or raise an issue on GitHub, if you're having problems 🛠️.