Real-time traffic quantization using a mini edge artificial intelligence platform

Abstract

Traffic analysis is dependent on reliable and accurate datasets that quantify the vehicle composition, speed and traffic density over a long period of time. The utilisation of big data is required if equitable and efficient transportation networks are to be realised for smart, interconnected cities of the future. The rapid and widespread adoption of digital twins, IoT (Internet of Things), artificial intelligence and mini edge computing technologies serve as the catalyst to rapidly develop and deploy smart systems for real-time data acquisition of traffic in and around urban and metropolitan areas. This paper presents a proof of concept of a mini edge computing platform for real-time edge processing, which serves as a digital twin of a multi-lane freeway located in Pretoria, South Africa. Video data acquired from an Unmanned Aerial Vehicle (UAV) is processed using a neural network architecture designed for real-time object detection tracking of vehicles. The implementation successfully counted vehicles (cars and trucks) together with an estimation of the speed of each detected vehicle. These results compare favourably to the ground truth data with vehicle counting accuracies of 5% realised. Detection of sparse motorcycles and pedestrians were less than optimal. This proof of concept can be easily scaled and deployed over a wide geographic area. Integration of these cyber-physical assets can be incorporated into existing video monitoring systems or fused with optical sensors as a single data acquisition system.