Learning to see: A computer vision story, Hristo Avramov, Data Science Team Lead at Ocado

At Ocado, we call our grid of storage containers “cells.” As bots whizz across the top of them at great speed, we refer to the platform as “the Hive.” It’s the world’s densest automated storage system, as far as we know. CCTV cameras provide continuous visual monitoring of the Hive, allowing us to identify each bot at any given time.

To make the most of this visual monitoring data, it’s essential to map camera images accurately to the grid coordinates. This is where the camera calibration algorithm comes in—converting image pixels into real-world grid positions and supporting all our CCTV projects.

In this talk, Hristo shares the story of developing and implementing this crucial algorithm. He discusses the research behind it, how it was brought into production, and how its performance was evaluated. Hristo also talks about the challenges faced, such as adapting to new CFC designs, different camera types, and various grid configurations. Through this, he aims to share valuable experiences and lessons learned.

***

techBytes is a series of events hosted by Ocado Technology across different countries and focuses on exchanging experiences in the area of engineering practices.