A Real Time Object Detection System Using a Webcam with Yolo Algorithm

Abstract

The main objective of object detection is to detect various objects in a video stream with more accuracy and with less computation time. Image classification involves assigning a class label to an image, whereas object localization involves drawing a bounding box around one or more objects in an image. The object detection algorithm YOLO to examine the entire image in a single instance and predicts the bounding box coordinates and class probabilities for these boxes. This paper uses Yolo algorithm with Darkent architecture to detect all possible objects in a video stream derived from a Webcam. From the experimental study, the accuracy of object detection varies from 37% to 88%. The existing algorithms take more time to process each frame and accuracy is also less. The Intersection over Union will decide prediction of each object as Good one. The Nonmax suppression technique will get a single prediction per object. The biggest advantage of using YOLOv3 can process 67 FPS. The experiment was conducted in a Colab Google environment with GPU. The video taken from various places were used to examine the effectiveness of the proposed approach. The experimental results show that our proposed approach can effectively improve the accuracy rate in detection of various objects in an image.