Safe distancing is important now as we fight Covid19, it is important also to help us return as much as possible back to normalcy. Being able to continually and cost effectively monitor safe distancing at crowded locations and being able to quantify and respond to it real time will help

We currently are only able to deploy a limited amount of safe distancing ambassadors. Due to limitations in manpower, they cover crowded and central markets/areas mainly. Delivery riders are hard to manage, stall owners pull their masks down and gather once ambassadors are gone, markets are cramped and it is hard for popular stalls to maintain safe distancing.

Policy makers may be hesitant to reopen up businesses and locations confidently as its hard to monitor and manage preventative behaviour as well as crowd levels in so many locations. Being able to quantify that in near time helps planning and response, as well as effective deployment of resources.

What it does

  • A network of lightweight IoT edge smart cameras that use computer vision to monitor and calculate safe distance between people
  • This can be used to monitor a queue or be used to monitor potential crowded spaces like a wet market.
  • Using capabilities already built into mobilenet v2 and running on a lightweight single board computer with a cheap camera attachment, we can estimate distance between 2 humans.
  • If it continually detects people close to each other for an extended period (x minutes) it can do a few things
    • play an audio alert (like in MRT trains) to remind the public to maintain a safe distance
    • send an alert to the attending safe distancing ambassador to inform them of the breach in safe distancing and the specific area it is happening. This reduces the need for safe distancing ambassadors to walk around or sit around constantly eyeballing. Also reduces the number of ambassadors required.
      • record this as an issue in a database which can be aggregated to eventually produce a safe distancing score for a location/stall/area in near time.
  • Quantified safe distancing data can be used in a number of ways:
    • Can be part of a safe distancing weight/score for an area/location/stall
    • A real/near time pulse on the number of safe distancing breaches so that ambassadors/officers can be deployed more effectively
    • Shared with various public so they can decide which areas are generally safer for them to visit
    • Can be used for backward analysis
    • Can be used for forward response of potential clusters/dangerous behaviour
    • The scoring system can be a way to help policy makers open up and monitor businesses and the economy more confidently

How we intend to build it

  • we have already assembled a half built version of it using mobilenet v2, a single board computer and a camera. It can already calibrate for warp and estimate distance between people.
  • data feed can be piped into a datawarehouse like bigquery using fluentd
  • data can be reported using SQL and Data-Studio for agility
  • alerts can start with e-mail but can be extended to telegram/slack/whatsapp

What's next for automated safe distance monitor

  • This can be combined with other add ons to monitor occupancy to form a more holistic scoring system
  • This can be combined with other add ons to monitor safe mask wearing to form a more holistic scoring system
  • This can be deployed in supermarkets
  • This can be deployed in shopping malls

Built With

Share this project: