Proposed an edge computing device design (IP67-rated, vehicle-mounted) for georeferenced image capture, addressing Bolivia's manual inspection bottleneck (5km/day vs. continuous automated coverage).
Designed multi-model architecture: semantic segmentation (U-Net) for crack detection, object detection (YOLO) for potholes, and lane marking segmentation—classifying severity and type per AASHTO standards.
Planned custom dataset curation of 1000+ images capturing Andean pavement conditions, addressing regional data gaps in standard open-source datasets.
Architected cloud-based REST API and dashboard design for asynchronous processing, generating standardized reports with Pavement Condition Index (PCI) for road management agencies.
Proposed novel integration of pavement distress and lane marking evaluation in a single system—a research contribution not previously explored in the literature.