Challenge
A client from the construction industry needed a dataset for automatic detection of construction equipment on worksites. The goal was to annotate all machinery in the images to enable automated monitoring and tracking of equipment movement.
Key Objectives
- Image Processing: Annotate construction equipment using object detection techniques.
- Dataset Creation: Develop a labeled dataset with equipment classes, including dump trucks, cement mixers, and cranes.
Solution
Task Definition & Technical Requirements
The client provided a list of equipment classes to be annotated, along with detailed technical specifications outlining precise labeling instructions. After carefully reviewing the requirements and clarifying details, we began the annotation process.
Efficient Annotation
Thanks to a well-prepared workflow, we completed the annotations in just five days. Clear technical guidelines and pre-approved object classes allowed us to streamline the process. The bulk of the annotation work was finished in four days, and when the client provided additional images, our team quickly processed them as well.
In total, we labeled 20,000 bounding boxes for various types of construction equipment across different locations and conditions. Each equipment class was meticulously annotated according to the given specifications.
Validation Process
To ensure maximum accuracy, all data underwent an additional validation stage. This involved selecting a representative sample of images for quality checks.
During validation, we proactively communicated with teams, reporting detected anomalies and providing insights on top-performing and underperforming annotators to team leads.
Training & Quality Assurance
We place strong emphasis on the quality of our validators' work. Their ongoing skill development is overseen by a dedicated training department.
| Stage | Input | Workflow Scope | Main Quality Checks |
|---|---|---|---|
| Task Definition & Setup | Equipment list, technical specifications | Review requirements, clarify details, prepare annotation workflow | Requirement clarity / Class coverage |
| Annotator Training & Pilot | Guidelines, reference images | Train annotators, conduct pilot annotations, align interpretations | Annotation accuracy / Pilot validation |
| Full Annotation | Construction site images | Annotate machinery using object detection, label bounding boxes | Label correctness / Completeness of annotations |
| Validation | Annotated images | Sample review, anomaly reporting, validator performance monitoring | Accuracy / Guideline compliance |
| Final Delivery | Validated dataset | Consolidation, final QA, submission to client | Dataset completeness / Usability for AI models |
The Results
- Timely Completion: All images were annotated within five days, including client-requested revisions.
- High Efficiency: The project was completed with exceptional accuracy and speed, allowing the client to utilize the data for real-time equipment monitoring.
- Positive Feedback: The client was highly satisfied with the quality and timeliness of the work, noting that all requirements were fully met.
Accurate object detection starts with precise labeling: clear classes, detailed guidelines, and a skilled annotation team turn raw images into actionable data.
- Roman Lukoshin
- Speech and Generative Data Manager
