Vehicle queuing at complex intersection scenarios via mmWave radar-camera fusion

Abstract

The queuing effect of vehicles perceived by road-side sensing equipment is very important which will directly affect the control quality of the signal. We proposed a vehicle queuing algorithm via mmWave radar-camera fusion to optimize queuing in complex intersection scenarios. Three types of stopped state was introduced into the tracker, which are FirstStop, VisionStop and LostStop. Shaded area caused by the vehicles was calculated and the lost target in the shaded area would be predicted along the lane. The shaking phenomenon of the stopped vehicle would be eliminated, the queue vacancies would be alleviated, and the timely location update would be realized when the stopped vehicle restarts. Through extensive evaluations of real-world datasets, the proposed method demonstrates remarkable improvements in the accuracy of queue length, which is more than 90%, and the ID switch of the stopped target is reduced by about 3 times/vehicle.