A Novel Framework for Real-time Traffic Flow Parameter Estimation from Aerial Videos

Abstract

Unmanned aerial vehicles (UAVs) are gaining popularity in traffic monitoring due to their low cost, high flexibility, and wide view range. Traffic flow parameters such as speed, density, and volume extracted from UAV-based traffic video are critical for traffic state estimation and traffic control, and has recently received more and more attention from researchers. However, different from stationary surveillance videos, the camera platforms move with UAVs and the motion in aerial videos makes it very challenging to process for data extraction. To address this problem, a novel framework composed of two complementary approaches for real-time traffic flow parameter estimation from aerial videos is proposed. The first approach is a motion-based approach, which identifies traffic streams and video background based on their motions using Kanade-Lucas-Tomasi (KLT) tracker and k-means clustering algorithm, and then extracts traffic flow parameters (speed, density, and volume) using connected graph and traffic flow theory. The second approach is a detection-based approach, which requires a vehicle detector training process. In this approach, vehicles from a top-view perspective are detected by the vehicle detector first and then the vehicle motion is estimated using KLT tracker as well as the background motion. Specifically, the vehicle detector is a combined cascaded classifier composed of Haar-like features and neural networks, making use of the fast processing speed of cascaded Haar classifier and the high detection rate of neural network. These two complementary approaches have their own advantages and together form the proposed framework for aerial video-based traffic flow parameter estimation. The system was tested on multiple aerial videos taken by UAVs operated in various scenarios including uncongested traffic condition, uncongested traffic condition, daytime, nighttime, UAV moving and UAV hovering. The experimental results show that the system is able to extract traffic flow speed, density and volume, and also achieves high performance in both traffic speed and vehicle count estimation in various challenging scenarios. The proposed system achieves a fast processing speed that enables real-time traffic information estimation.