Performance Analysis of Object Detection Framework: Evolution from SIFT to Mask R - CNN
Main Article Content
In a near of wide spread technological change that has givena positive impact to the society andhelpedinbuildingauser-friendly environment, object detection framework, an importantpart of Computer Vision (CV) plays a vital role. Starting fromasimpleautomaticattendancesystemforstudentsusingfacedetection, recognizing the presence of tumors in medical images,helping with automatic surveillance of cctv cameras to identifypeoplewhobreakstrafficrulescausingroadaccidentstobeingthecentral mechanism behind self-driving cars, object detection haswide range of applications and assist building an easy to cope withsmart environments. This in turn urges the need to evaluate theperformanceofthetechniquesbehindtheseframeworks.Thecentralideabehindthemodern-dayobjectdetectionandclassification is Convolutional Neural Network (CNN) which triesto mimic the occipital lobe, the visual cortex of the human brain.CNNhaswiderangeofvariationsandhascomethroughalongwaystarting from basic CV techniques like Scale Invariant FeatureTransform(SIFT),HistogramsofOrientedGradientsn(HOG)tillRegionbasedCNN’s(R-CNN).Theperformanceofeachandeverymethodthathas led throughthe evolutionofobjectdetectionmethods, its advantages and the disadvantages which has pavedway for the innovation of next technique has been discussed andrepresentedindetail.