Real Time Vehicle Detection and Counting Using Deep Neural Network

Real Time Vehicle Detection and Counting Using Deep Neural Network

Shilpa Yadav, Shakshi Verma, Prof. Ashish Baiswar

Department of Information Technology,

Shri Ramswaroop Memorial College of Engineering and Management

Abstract

The number of vehicles on the road grows daily. Real-time vehicle counting can efficiently improve traffic  management system Aiming at efficiently collecting traffic information in real time, the author proposes an effective vehicle counting system for detection and. tracking vehicles in complex traffic scenes. This article presents a template for detecting different vehicles and counting them within a given video frame.. Using YOLOv4, we can detect the various types of vehicles in the given video.

Additionally, it uses the DeepSORT algorithm to help count the number of vehicles going through the video efficiently. Performance parameters such as iou and map are calculated in order to measure the performance and validation of the work. After implementing the algorithms, Based on the YOLOv4 test results yielded a detection accuracy rate with mAP of 84.50%. Where YOLOv4's combination with the in-depth sorting algorithm can detect, track and count four vehicle types.

Keywords: YOLOv4, DeepSort, mAP ,Tracking