Computer Vision Engineer — Construction Site Intelligence

Mia Johansson

Freelance · Mid-level · London

This role is no longer accepting applications

Similar open roles you can apply to now:

About the role

We build software for large-scale infrastructure and commercial construction projects. Our platform connects BIM models, site photography from weekly drone surveys and fixed cameras, and IoT sensor data from concrete maturity sensors and structural monitors to give project managers and site engineers a unified view of programme progress. The computer vision piece of our platform — progress detection from site photography — is currently powered by a vendor API that we have been using since launch. It works for straightforward structural progress detection. It does not work for the things our clients have started asking for: formwork identification, rebar density estimation, safety PPE compliance screening, and activity classification from time-lapse sequences. We are commissioning an eight-week project to build prototype models for at least two of these detection tasks, evaluated on our own dataset of annotated construction site imagery. We have 340,000 annotated images across 22 active project sites. If the prototypes meet our accuracy thresholds, we will integrate them into the platform and look for an ongoing engineering relationship.

Key Deliverables

List the expected deliverables for this project.

Select and scope two target detection tasks from the client list in consultation with the product and engineering teams
Build training pipelines from our annotated construction site imagery for the selected tasks
Train and evaluate detection models to agreed accuracy thresholds — produce a written evaluation report with clear recommendations
Optimise inference performance for deployment in our existing containerised serving infrastructure
Conduct a structured handover with the engineering team including documented architecture, training pipeline, and evaluation methodology

Requirements

Technical stack needed for this mission.

4+ years of computer vision engineering with experience building object detection or classification models on real-world, non-benchmark datasets
PyTorch for training and evaluating detection models — YOLOv8 or equivalent for real-time inference tasks
Python for data pipeline development, annotation tooling integration, and model evaluation
Computer Vision beyond classification: object detection, semantic segmentation, and the practical challenges of outdoor, variable-lighting, occluded imagery
Scikit-learn for baseline model analysis and evaluation metric calculation
Docker for reproducible training and evaluation environments that our engineering team can build on after the engagement
Familiarity with construction, BIM, or AEC domain is an advantage but not a requirement — a willingness to understand what the detections are used for is