Automated safety scoring for points of interest

Inspiration

While we are the current stage of tackling the spread of Covid-19, our country has already been looking ahead to ensure we are well positioned for recovery, to be able to open up our domestic economy quickly and safely. There is however a constant worry of recurrent waves of infections that we must contain. In order to reopen our economy there will be new norms to adhere to.

Given the current measures we have put in place, we do not have enough manpower to fully manage and ensure various locations are relatively safe. It is hard to quantify and consolidate currently manually observed data in a timely manner.

A suite of solutions can be put together to create a safety score/weight which can be transmitted near time to a centralised source where various dashboards, monitoring and alerts solutions, and various agencies can access to be able to better manage the re-opening up of the economy.

What it does

• Uses low cost computer visions solutions to monitor social distancing in queues and within locations like markets, transport hubs, malls...
• Uses low cost computer vision solutions or wireless packet sniffing solutions or online crowd data sources to measure occupancy in a location like a wet market, a building, supermarkets, transport hubs...
• Uses low cost computer vision solutions to check if individuals are wearing masks and even if individuals are wearing masks correctly.
• These systems can run 24/7 and monitor the behaviour of essential service staff to prevent gatherings and the lax usage of PPEs.
• The data from each measure will be fed to a centralise data warehouse near time which will then calculate a score for each point of interest.
• This data can be used to map out where potential hotspots might be in which can then trigger corrective action
• The data can also be shared in some form publicly much like spaceout.gov.sg to help the public plan their journey or destinations

How it can be built

• Computer visions solutions can be built on single board computers like odroid with an attached low cost camera using mobilenet v2 or a similar machine learning library. I have a team that has already been experimenting with this.
• Data can be piped into a cloud service using a pub sub messaging infrastructure like kafka or fluentd
• Data can be warehoused in cloud solutions like redshift or bigquery
• Alerts and triggers can be built in python services and dashboards can be generated in tableau, data studio or powerbi depending.

Potential challenges

• There are some hardware costs
• There are real deployment issues like networking, mounting and power supply

Built With

• bigquery
• kafka
• leaflet.js
• maptive
• mobilenet
• mqtt
• odroid
• python
• sql