Problem Statement:
A client had a unique challenge: they needed to know their inventory levels on their shop floor at each of their facilities, but could not allocate manpower for manual counts.
Solution:
Using consumer-grade webcams, Python, and Support Vector Machines, we were able to build a machine learning toolkit that determined daily inventory levels to a high degree of accuracy.
Complication:
While machine learning tools typically require high numbers of observations for classification problems (especially when working with images), we were able to expand our observation counts using light modifications of tones, colors, windows, and jitters that maximized learning opportunities without sacrificing real-world accuracy.
Outcome:
We deployed a nightly classification routine that grabs the latest image from the shop floor, applies the classification model, and updates a database and also notifies the team daily with a Slack message.