Implementation of YOLOv4 to Perform Vehicles Detection and Counting Using Aerial Images of an Open Car Park

Goh, Hui Siang (2021) Implementation of YOLOv4 to Perform Vehicles Detection and Counting Using Aerial Images of an Open Car Park. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text GOH HUI SIANG _Full text.pdf Restricted to Registered users only Download (6MB)

Abstract

Students in TAR UC have been endured in finding a parking lot in the TAR UC campus due to the number of students who have greatly outnumbered the number of parking lots. Due to the portability, adaptability, and the capacity of covering large areas, unmanned aerial vehicles (UAV) are being utilized in traffic surveillance. The idea of using UAV as car park management system in TAR UC has been brought up. However, detecting and counting cars in aerial images are challenging due to cars in aerial images are smaller in scales. Cars’ appearances and the complex background of images also affect the detection. In this project, it is aimed to develop an approach that able to identify and count the cars through aerial images by using YOLO (You Only Look Once) v4 algorithm. A custom dataset contains of aerial images with cars are prepared to train and test a model that combine YOLOv4 and TensorFlow. The YOLOv4 + TensorFlow custom detector is able to detect and count the cars in the aerial images. The model is then compared and outperformed the other state of the art methods by achieving a recall score of 100% and a F1-score of 0.99.