In collaboration with an infrastructure company based in Belgium, I developed a computer vision pipeline to automatically detect and localize key components on a worksite in France.
Using a custom-trained YOLOv11 model I was able to achieve high-precision object detection across a variety of industrial elements, including pipelines, valves, and structural fixtures — all captured directly from field photographs. The dataset consisted of real worksite images, annotated manually, with the goal of automating inspection and documentation processes.
Once the model was trained and validated, I implemented an inference engine capable of processing new site photos and overlaying bounding boxes with confidence scores on each detected component. This output was then used to automatically generate a schematic representation of the site layout, offering a high-level overview of the structure without manual CAD work.
The entire pipeline, from training and inference to the final schema generation — is fully automated.
This project involved:
- Dataset preparation and annotation
- YOLOv11 training and hyperparameter tuning
- Real-time inference and post-processing
- Schematic generation from detection coordinates
- Integration with external tools for downstream analysis
This system helps reduce manual labor and improves the consistency of site documentation.