In my role at the company I work for I, alongside my team, are responsible for suggesting and implementing innovative new systems which make us stand out from other companies in our industry.
One of the systems we are implementing right now is Machine Learning (ML), where ML is a subset of Artificial Intelligence. 1
Our site operatives, using our bespoke software, often have to carry out data collection jobs for local authorities, mainly on street furniture (street lights, bollards, signs etc.) with upwards of 80 pieces of information available for each asset.
With there often being around 50,000 assets to collect data on, with upwards of 80 fields, with up to 10 different operatives on site possibly interpreting things differently, you get two distinct problems:
- Inconsistency of data
- Data entry errors
To try and minimise these issues we are creating ML models for being able to pick out key elements of data from the photographs our operatives take on site, using the extensive photo library we have from the past 15 years.
The office trials we have run have been successful to say the least, with a 95% accuracy of prediction on 3 different fields of data so far, so given the time to train we believe we can increase the consistency and reduce the errors.
I gave a 1 hour presentation to one of our larger clients in 2019, and it seemed to be very well received. We even did a live demonstration. At the start of the presentation I asked the audience to email me pictures of street lights, so we could demonstrate identifying the material and the bracket type. I got some very interesting images, some of which were obviously sent to try and catch us out, but we passed with flying colours.