This project implements a boat detection system developed as part of a Computer Vision course. According to the assignment description, the system was required to be built without using end-to-end state-of-the-art detectors such as YOLO.
The system follows a two-stage detection pipeline:
-
Region Proposal Utilizes OpenCV’s Structured Edge Detection and EdgeBoxes to efficiently generate candidate regions of interest.
-
Classification A Convolutional Neural Network (CNN) processes the proposed regions to classify whether they contain a boat.
Two-stage design: Region proposal + CNN classification
Non-Maximum Suppression (NMS) for duplicate removal
IoU-based evaluation metrics for quantitative performance assessment