Innovative Vehicle Location Tracking and Traffic Analysis using Aerial Imagery Combined with Finite Basis Physics-Informed Neural Networks for Enhanced Urban Traffic Management

Abstract

Vehicle location tracking is the process of using tracking equipment to monitor a vehicle's actual time geographical position. It allows fleet managers, organisations, and people to monitor vehicle movements, optimise routes, and assure safety. As these systems continuously monitor and record vehicle locations, they may accidentally violate individual's privacy by tracking their movements without their knowledge or agreement. In this manuscript, to integrate aerial imagery and Physics-Informed Neural Networks for vehicle tracking and traffic management (VLT-TAI-FBPINN-UTM) is suggested. First, input images are collected from Vehicle Detection in Aerial Imagery (VEDAI) dataset. In order to do this, the incoming image is pre-processed using the Bilinear Double-Order Filter (BDOF), which is used to denoise and resize the image. Then the pre-processed images are fed into Fast Continual Multi View Clustering to segment the different sector of the aerial images. The segmented images which are fed into Finite Basis Physics-Informed Neural Networks (FBPINN) to detect and track the locations of various vehicles in the aerial images, as well as classify the vehicles into categories for example, a truck, a pickup, a tractor, a van, a car, a camper, a plane, a boat, and so on. In general, FBPINN does not incorporate adaptive optimization strategies to determine optimal factors for ensuring accurate detection of various vehicle locations in aerial images. Then the proposed VLT-TAI-FBPINN-UTM is presented in Python and the response such as Accuracy, Precision, Recall, F1 Score and Specificity are analysed. Result analysis of the VLT-TAI-FBPINN-UTM approach attains 29.5%, 27.23% and 21.14% higher accuracy; 17.9%, 21.6% and 29.45% greater Precision and 14.5%, 20.63% and 19.15% greater Recall when analysed through existing techniques such as smart traffic monitoring using pyramid pooling vehicle detection and filter-based tracking on aerial images (STM-VD-TAI-CNN), improving vehicle tracking and detection in UAV imagery using a particle filter approach and pixel labelling (VDT-UAV-YOLOv3), and a new framework for vehicle tracking and detection in nightware surveillance systems (VDT-NWSS-SVM) methods, respectively.