Real Time Object Detection

EasyChair Preprint no. 13354

Abstract

This work proposes a versatile improvement of the well-known object detection procedure called Region-based Convolutional Neural Networks (R-CNN). Unlike the previous methods which achieved impressive speed but low accuracy, YOLO can be considered in a similar category. On contrary, R-CNN involve multi-stage pipeline that are such as region proposal generation, feature extraction, and classification to achieve greater object localization accuracy. R-CNN utilizes this method to outperform YOLO in MAP scores which are an important measure of detection accuracy. Even though it imposes high computation demand, R-CNN proves to be more promising in terms of false positives, especially on complex backgrounds, which make it a more appropriate approach for a range of applications. Interestingly, R-CNN is not only stable, but also works better than YOLO and the other latest approaches when we need to identify objects in different domains, such as paintings and natural scenes. And this is the most significant project, which consist of both the real-time as well as the capture image to detect the object and multi object detection is also working properly by using the R-CNN model.

Keyphrases: Convolutional Neural Networks, object detection, Region-based Convolutional Neural Networks

@Booklet{EasyChair:13354,
  author = {Faiayaz Waris Saiyed and Narasimha Reddy Anumula and Mahesh Babu Elchuri},
  title = {Real Time Object Detection},
  howpublished = {EasyChair Preprint no. 13354},

  year = {EasyChair, 2024}}