So many things to learn

December 4, 2020 1 minute read

I was not making any records for almost 2 months. Will do that now.

Progress on VideoParking

There was quite some progress on the VideoParking.live project:

We have now a Raspberry PIs with cameras, at 2 (almost 3) locations, that upload a photo every minute to S3
There is a Lambda triggered by such uploads, that produces zones picture (initially it was producing a zip file with each zone as a separarte jpg file, but it showed that running detection on a file has a constant time and it’s around 45 seconds on home server, so no zones are just masked on one jpg) and uploads it to different S3 bucket
Then there is a detector processes, running in k8s, sharded, which fetches zones file for its locations every minute, performs detections and registers results to Timestream
On the page, react app is fetching data (location stats per zone, raw image, and detections) and drawing everything on a canvas.

Plan is to show only stats on a map and add authorization for billing and operators’ features (raw image and all detections data).

react-leaflet: to display zones and parking status, show current location, find locations in the area, cluster zones.
react-leaflet-draw: to draw zones on the map easily.
Prometheus and Grafana for k8s cluster: to collect metrics from services.
Drawing on canvas in React App: to ease zones corrections/creation on raw image.
Authentication for React app and Lambda functions.
YOLO3 detector: how to train it on for cars only and to detect cars from the top view angle.
Running detector in EKS and probably on GKS, AKS to compare.
Enabling GPU support on local k8s cluster and using image with GPU in cloud.

Yeah, the race started at the first part of November. And I attended RYA skipper theory courses. Big plans.

And yesterday made an account in Virtual Regatta, now I’m also doing round the globe.

You can. You should. And if you’re brave enough to start, you will. —Stephen King

And T –(+W)–> C.