A Deep Vision and UAV Paradigm for Future Intelligent Mobility

Abstract

These days, the advancement of Information and Communication Technology (ICT) has affected our daily lives significantly, including governance. Urban public infrastructure facilities including Transportation Systems, Water and Sewer networks, and energy supplies are getting intelligent using open-source data-driven platforms in the modern world. Unfortunately, due to the lack of resources and poor governance in underdeveloped countries, the local circumstances of the cities like Karachi have not been extensively explored by researcher's despite of growing population. This paper mainly discusses the viability of using computer vision and unmanned aerial vehicles in object detection for congested roads in Karachi. The famous CNN algorithms like YOLOv5 and Cascade classifier are modified and used for vehicles running on the overcrowded road for accurate recognition, counting, and speed anomaly detection. Moving further, the License Plates (LP) have been recognized using Optical Character Recognition specifically considering the local parameters. A comparison of accuracy, robustness in different surroundings, and environments using Cascade Classifier, and Tesseract OCR has been performed. Custom model training on YOLOv5 has been performed and an 'Accuracy Enhancement' technique integrating real-time drone stream data with YOLOv5 and easy OCR using a mean averaging algorithm has been proposed to get better accuracy in license plate detection.