Deep Learning based Object Detection using Mask RCNN
Triphena Delight D, V. Karunakaran
Abstract
Object detection aims to recognize all instances of a known class of objects in an image, such as people, vehicles, or faces. In recent times deep learning techniques was applied for detecting objects and those existing methods have some problems with viewpoint changes and occlusion. This paper proposes a solution for automatic object detection by implementing instance segmentation at a pixel level, and several RCNN techniques were also inferred in this paper. The proposed Mask RCNN detects the images and marks them with bounding boxes, class labels as well as masks. The implemented mask RCNN model was trained and tested with Plasmodium vivax dataset so that the model is capable to detect cells in a congested image. The model was also tested in a custom dataset to check the model performance, and the model obtained 94% mAP for the custom dataset. This paper also mentions the future scope of this work so that the robustness and the reliability of the model can be improved further.