A Deep Reinforcement Learning Approach for Ramp Metering Based on Traffic Video Data

Ramp metering that uses traffic signals to regulate vehicle flows from the on-ramps has been widely implemented to improve vehicle mobility of the freeway. Previous studies generally update signal timings in real-time based on predefined traffic measurements collected by point detectors, such as traffic volumes and occupancies. Comparing with point detectors, traffic cameras—which have been increasingly deployed on road networks—could cover larger areas and provide more detailed traffic information. In this work, we propose a deep reinforcement learning (DRL) method to explore the potential of traffic video data in improving the efficiency of ramp metering. Vehicle locations are extracted from the traffic video frames and are reformed as position matrices. The proposed method takes the preprocessed video data as inputs and learns the optimal control strategies directly from the high-dimensional inputs. A series of simulation experiments based on real-world traffic data are conducted to evaluate the proposed approach. The results demonstrate that, in comparison with a state-of-the-practice method, the proposed DRL method results in (1) lower travel times in the mainline, (2) shorter vehicle queues at the on-ramp, and (3) higher traffic flows downstream of the merging area. The results suggest that the proposed method is able to extract useful information from the video data for better ramp metering controls.

Enhanced Spatial–Temporal Map-Based Video Analytic Platform and Its Local- Versus Cloud-Based Deployment with Regional 511 Camera Network

This paper explores the cloud- versus server-based deployment scenarios of an enhanced computer vision platform for potential deployment on low-resolution 511 traffic video streams. An existing computer vision algorithm based on a spatial–temporal map and designed for high-angle traffic video like that of NGSIM (Next Generation SIMulation) is enhanced for roadside CCTV traffic camera angles. Because of the lower visual angle, determining the directions, splitting vehicles from occlusions, and identifying lane changes become difficult. A motion-flow-based direction determination method, a bisection occlusion detection and splitting algorithm, and a lane-change tracking method are proposed. The model evaluation is conducted by using videos from multiple cameras from the New Jersey Department of Transportation’s 511 traffic video surveillance system. The results show promising performance in both accuracy and computational efficiency for potential large-scale cloud deployment. The cost analysis reveals that at the current pricing model of cloud computing, the cloud-based deployment is more convenient and cost-effective for an on-demand network assessment. In contrast, the dedicated-server-based deployment is more economical for long-term traffic detection deployment.

Detection and Tracking of Lane Crossing Vehicles in Traffic Video for Abnormality Analysis

In this paper, we present a novel approach for detection and tracking of lane crossing/illegal lane crossing vehicles in traffic video of urban highways. For that intention, an initial pace is performed that estimates the road region of the geometrical structure. After finding the road region, every vehicle is tracked in order to detect lane crossing vehicles according to the distance between lane lines and vehicle centre, it is followed by tracking of lane crossing vehicles based on model-based strategy. The proposed system has been evaluated using recall and precision metric, which are received using experiments carried on selected video sequences of GRAM-RTM dataset and publically available video sequences. The experimental results present that our method reaches the highest accuracy for detection of vehicles and tracking of lane crossing vehicles.

Road Traffic Congestion (TraCo) Estimation Using Multi-Layer Continuous Virtual Loop (MCVL)

Any road traffic management application of intelligent transportation systems (ITS) requires traffic characteristics data such as vehicle density, speed, etc. This paper proposes a robust and novel vehicle detection framework known as multi-layer continuous virtual loop (MCVL) that uses computer vision techniques on road traffic video to estimate traffic characteristics. Estimations of traffic data such as speed, area occupancy and an exclusive spatial feature named as corner detail value (CDV) acquired using MCVL are proposed. Further, the estimation of traffic congestion (TraCo) level using these parameters is also presented. The performances of the entire framework and TraCo estimation are evaluated using several benchmark traffic video datasets and the results are presented. The results show that the improved accuracy in vehicle detection process using MCVL subsequently improves the precision of TraCo estimation. This also means that the proposed framework is well suited to applications that need traffic characteristics to update their traffic information system in real time.