Deep Learning Application for Vehicle Detection through Surveillance Drones

Vehicle detection plays an essential role in a variety of real-world applications, including autonomous driving, traffic surveillance, and management. This research work thoroughly compares the three deep learning models for vehicle detection— You Only Look Once (YOLOv8), Faster Region-Convolutional Neural Network (R-CNN) and Single Shot MultiBox Detector (SSD). The study is conducted using different datasets, especially the Pak Vehicles Dataset, which consists of diverse images taken in different environmental situations across Pakistan. It also includes a description of annotation methods and explains the architecture of each model. The evaluation focuses on accuracy, speed, and robustness metrics like Mean Average Precision (mAP), recall, and inference time. The result highlights strengths, weaknesses, and insights for improving real-world vehicle detection systems. For full implementation of these models, kindly check this: https://shorturl.at/nJRSV