Image-Based Learning to Measure the Stopped Delay in an Approach of a Signalized Intersection

Abstract

Traffic delays are inevitable when evaluating the performance of a signalized intersection, but these delays cannot be directly measured in the field based on existing spot detectors. Traffic-light controllers have adopted a reinforcement learning (RL) algorithm, which is currently prevalent in the field of study and requires real-time measurement of traffic delays to derive the state and reward for each time period. No RL-based study, however, has provided a robust way to measure traffic delays. In order to bridge the gap, we devised a convolutional neural network (CNN) to directly measure traffic delays from video footage in an end-to-end manner. The proposed methodology proved superior to both a state-of-the-art vision technology and an analytic formula that has widely been used to estimate delays. Furthermore, a robust method to secure labeled data without human input was suggested based on a cycle-consistent adversarial network (CycleGAN). INDEX TERMSTraffic delay estimation, image-based learning, deep convolutional neural network.

Area-wide traffic signal control based on a deep graph Q-Network (DGQN) trained in an asynchronous manner

“…However, there may be skepticism regarding how to accurately measure these parameters when an RL algorithm is implemented in the field. Our two previous studies showed strong evidence that they can be measured onsite based solely on video images (Shin et al, 2019;Chung and Sohn, 2017). The state was defined for each lane group that is comprised of one or more lanes that share a common stop-line and capacity.…”

Development of a Theoretical Delay Model for Heterogeneous and Less Lane-Disciplined Traffic Conditions

“…Researchers have also relied on dynamic approaches for delay estimation using state space modelling and filtering techniques [11][12][13][14]. Researchers have also used more dataintensive techniques like time series analysis, k-means clustering, and other artificial intelligence based solutions for delay estimation [15][16][17][18][19][20]. Researchers have also employed various metaheuristic approaches like genetic algorithm, differential evolution, harmony search, and artificial bee colony for estimating and optimising intersection delay [21][22][23].…”

Measuring Traffic Volumes Using an Autoencoder with No Need to Tag Images with Labels

“…Furthermore, YOLO requires additional training whenever the testbed changes. Drawing a bounding box for every vehicle in training images entails considerable human effort, but it is required in order to accurately recognize vehicles in photos taken from different viewpoints [22].…”

“…To tackle the problem, original images were simplified using a CycleGAN developed by [23], so that both vehicles and backgrounds would have monotone colors. Such a transformation had already been successfully adopted in our previous studies measuring traffic speed and delay [22,24].…”

“…This capability of a CycleGAN also facilitates traffic surveillance that depends on images. We have already used images synthesized by a CycleGAN to enhance the accuracy of measuring both traffic speeds and delays [22,24]. This scheme was adopted in the present study to obtain images that were more consistent at recognizing the presence of vehicles on the detection line.…”