Object Regognition and Detection using Yolo V3

Abstract

Object detection and identification is an entrancing field, and as such, apart from research applications automatic object detection applications are being witnessed in real-time domains such as non-stop surveillance and in different industries and businesses. The improved capabilities in both hardware and software have led to fast and pivotal discoveries in this arena. Convolutional neural network (CNN) is the most representative model of deep learning and in this paper, a variant of CNN namely  YOLO v3 is experimented with for localizing twenty different object classes that include aeroplane, person, car etc. The work is done using darknet-53 pre-trained model as the backbone network using open CV python, and TensorFlow 2.0. Images taken from two different datasets namely COCO and Pascal VOC datasets are given as input to the model and theoutput  in the  form of bounding boxes, accompanied by objectness score  and class label. The results indicate that YOLO V3 is very efficient in object detection.