Virtual Booth

Reliable perception of water surface obstacles is critical for maritime safety and collision avoidance in small boats. This project introduces WAIDAR, an edge-enabled stereo vision AI system designed for real-time surface obstacle awareness. WAIDAR utilizes an NVIDIA Jetson Orin NX with a fine-tuned YOLOv11 model trained on a custom Roboflow dataset comprising four classes (boat, debris, structures, aquatic plant). On unseen samples, the model achieved [email protected] of 0.7561, precision 0.7988, recall 0.6586 and F1-score 0.7220, demonstrating viable onboard performance. The prototype highlights the potential of WAIDAR for operational deployment in riverine environments and small vessel navigation. Ongoing work includes dataset scaling, stereo depth integration, and sensor fusion for improved safety.