Data Labeling and Annotation for AI Training
The project primarily focused on video footage and images captured by an ego car equipped with cameras mounted on the front, back, or both sides. This setup was designed to capture a range of environments, including urban, rural, highway, and tunnel settings, to aid the autonomous vehicle in real-time object detection and scene understanding. I collaborated with a team of 300 annotators and 10 Quality Analysts, working on over a million+ images and videos, totaling more than 600 hours of annotation. The task involved creating polygons and cuboids around various objects, such as buildings, lane markings, pedestrians, vehicles, traffic signs, and natural elements. Each label was assigned a different color for easy differentiation. We also labelled occluded or partially visible objects to enhance the model's robustness. Consistent and accurate tracking of objects across video frames was ensured. Additionally, we implemented a QA SLA of 98%.