RCNN Object Classification and Semantic Segmentation

Abstract

newline KEYWORDS: Multi-Object Detection; Object Detection; Feature Extraction; newlineSemantic Segmentation; Transfer Learning newlineApplications of computer vision, from autonomous vehicles to medical newlineimage analysis, rely heavily on object identification and classification. newlineObject detection is a computer vision technique that is used to find newlineoccurrences of objects in images or videos. With the introduction of deep newlinelearning, object identification algorithms have become much more precise. newlineIt allows a computer to simulate intelligence like a human who can quickly newlineidentify and pinpoint objects of interest when viewing photos or videos. newlineOne of the most important application of these is the development of newlineregion-based convolutional neural networks (R-CNN), which has made newlineit possible to recognise and categorise objects in images. The current newlinemethods of object classification and the depth of semantic comprehension newlineextracted from an image both have potential for development. newlineIn this thesis, the main focus is on making machine learning model for newlineobject detection that can be used to detect multiple objects in a given scene. newlineIt can be very useful for autonomous vehicles, surveillance and security newlinesystems, traffic management systems, pedestrian detection systems, animal newlinedetection systems, etc. newlineThis study improves semantic segmentation within the R-CNN newlineframework to tackle the problem of object classification. The three steps newlineof region proposal, feature extraction, and classification are carried out newlinesequentially in conventional R-CNN methods. For improved accuracy and newlinecontext-aware object classification, a novel integrated architecture that newlinecombines region-based object identification with semantic segmentation newlinewas developed. The addition of a Semantic Segmentation Module (SSM) newlineto the R-CNN pipeline is a major improvement brought about by this newline2 newlinestudy. The SSM improves feature extraction by giving more specific newlineinformation about objects at the pixel level. It creates object masks newlinedown to the pixel level, which are then utilised to h