An In-vehicle Camera Based Traffic Estimation in Smart Transportation

Abstract

Traffic condition estimation services a critical role on assisting vehicles to estimate current traffic states and determine an optimal route for their journey, thereby improving the traffic efficiency in smart transportation. To accurately estimate traffic conditions of roads, a number of efforts focusing on fixed-camera based traffic estimation has been developed. However, fixed-cameras in smart transportation usually are deployed on road-side units, leading to limited visual areas. Additionally, these existing schemes also are lack of flexibility and cannot achieve great traffic estimation efficiency to roads with less cameras deployed. To address these issues, in this paper an In-vehicle Camera based Traffic Estimation scheme (ICTE) is proposed for smart transportation system to assist vehicles to estimate traffic conditions with traffic videos collected by in-vehicle cameras. Integrating with edge computing and V2X (vehicle-to-vehicle (or infrastructure)) networks, our scheme can obtain extra computation and communication resource to accurately estimate traffic conditions anywhere vehicles reach. Particularly, YOLO-based vehicle detection and Hough-based lane detection are introduced in our ICTE to detect the vehicles and lanes in traffic videos captured by in-vehicle cameras. Then, a lane-based vehicle counting model is developed to count the number of vehicles that travels on different lanes. Finally, a density-based traffic estimation model is conducted to estimate traffic conditions based on the vehicle density. Via simulation experiments, our data shows that ICTE can achieve great accuracy and efficiency in terms of vehicle counting and traffic condition estimation.