A Novel Object Detection Framework using Convolutional Neural Networks (CNN) and RetinaNet

Abstract

Object detection is a computer vision technique that received high significant attention in recent decades. Object detection algorithms typically employ machine learning or deep learning to produce valid results. People can quickly recognize and locate objects of interest in the provided input images or videos. Object detection aims to use a computer to mimic this intelligence. Deep learning techniques have significantly improved the cure for object detection. This research aims to incorporate radical object detection techniques to achieve high accuracy. This research study covers a variety of factors and algorithms used in object detection methods using deep learning techniques detection algorithms, datasets and software hardware requirements used in the detection of objects. Benchmark datasets for object detection were discussed. The discussed method uses CNN with RetinaNet (Residual Networks and Feature Pyramid Networks) was implemented with benchmark dataset COCO. RetinaNet provides low loss and high accuracy (96%) Object Detection has a variety of applications like Autonomous driving, Traffic monitoring and Maintenance, People Counting and Video safety applications.