Pedestrian and Object Detection using Image Processing by YOLOv3 and YOLOv2

Abstract

Detecting instances of semantic objects of a specific class, such humans and other things, in digital photos and videos is the goal of object detection, a branch of computer vision and image processing. It is frequently used for activities including picture annotation, vehicle counting, and object tracking, such as tracing a ball during a football game, a cricket bat's movement, or primarily a person in a movie. In this study OpenCV is used with YOLOv3 neural network to detect pedestrians and objects from an input video or a real time webcam. In order to determine how accurately the pedestrians and objects are recognized using the You Only Look Once (YOLO) algorithm, a box is produced along its boundaries with its name and Intersection Over Union (IOU) value, which is determined using the formula area of Intersection/Area of Union. The pre-trained model and the weights of the COCO dataset of YOLOv3 tiny algorithm are used for the detection and compared with YOLOv2 algorithm.