Citywide reconstruction of traffic flow using the vehicle-mounted moving camera in the CARLA driving simulator

Estimating traffic flow is essential in planning road development, routing, navigation, autonomous driving, and other applications in road infrastructure management. Recently, cameras and portable devices mounted on moving vehicles have been proposed to estimate citywide traffic flow. Nevertheless, the efficacy of such algorithms has not been proven on a wide scale under varying environmental conditions. This paper modifies existing algorithms to estimate traffic flow from vehicle-mounted cameras in the driving simulator CARLA and conducts a comprehensive analysis of seven different traffic environments to assess the effectiveness of recently proposed techniques. The proposed algorithm achieves the highest accuracy, up to 84.2% evaluated using ground-truth data. With the proliferation of moving cameras (such as dashcams and smartphones) and inexpensive edge devices, the potential for the use of moving camera sensors in conjunction with edge devices for traffic flow estimation is promising. Our study provides more insight into how these techniques perform in different traffic environments.