Deep Learning based Object Detection using Mask RCNN

Object detection aims to recognize all instances of known classes of objects in an image, such as people, vehicles, or medical structures. Recently, deep learning techniques have been extensively applied for object detection; however, existing methods often encounter challenges such as viewpoint changes, occlusions, and fine-grained segmentation. This study proposes a solution for automatic object detection by implementing instance segmentation at the pixel level using a Mask R-CNN. The proposed system utilizes the Mask R-CNN framework to detect objects and delineate them with bounding boxes, class labels, and segmentation masks. The model was trained and evaluated on the COCO dataset, enabling it to address complex scenarios, such as occlusions and fine details. Furthermore, the system is integrated into a web application, providing a user interface for image upload, result visualization, and interaction. The proposed solution demonstrates high accuracy and scalability, rendering it suitable for healthcare, autonomous systems, and security applications. The model was additionally evaluated on a custom dataset to assess its performance, yielding a 94% mean Average Precision (mAP) for the custom dataset. Future research will encompass extending the system for video-based detection and custom dataset support to further enhance its robustness and reliability.