Vehicle Counting based on Convolution Neural Network

Abstract

In recent decades, intelligent transportation systems have drawn a lot of attention. For a traffic control strategy to be successful, accurate and thorough traffic flow information is essential. The main objective in this field is vehicle detection, and two crucial applications are vehicle counting and classification. There are numerous ways to count vehicles, including manual counts, computer vision, pneumatic road tube counting, and inductive loops. But developing a quick and precise approach for estimating traffic volume and vehicle counts is the major goal. The conventional approaches need a lot of time and complexity. Therefore, a computer vision-based virtual detection zone method that is straightforward, quick, and accurate is being used to address these problems. A UA-DETRAC dataset is used to evaluate the suggested approach. Firstly, the zone is made by setting the coordinates in the frame. This can be done by manually plotting the points on the frame. The zone is visualised into the frame using the OpenCV library. Pre-trained YOLOv3 model is used for object detection and classification. The vehicles are classified into five different categories. Sort algorithm is used for vehicle tracking and counting the number of vehicles that pass through the virtual detection zone. The number of vehicles passing through the virtual detection zone in a given time can be used to estimate the traffic volume at the time.