Joyoung Lee

Calibrating Communication Simulator for Connected Vehicle Applications

Research efforts focused on the connected vehicle (CV) technology applications typically assume perfect communications among the vehicles and between the vehicles and the roadside equipment. However, a few studies, including this one, pointed out that the wireless communications experience packet drops, which might lead to a serious downgrade of the safety critical CV applications. Thus, the wireless communication simulators used to emulate the communications performance need to be properly calibrated to replicate the real-world vehicular communication environments. This study calibrated an NCTUns simulator for the dedicated short range communications (DSRC) of CV technology using the DSRC field test data executed on an instrumented intersection at the Turner–Fairbank Highway Research Center. Physical layer parameters (e.g., data rate and transmission power) as well as channel models are calibrated. The calibration applied a Latin hypercube sampling technique to generate multiple combinations of parameter sets. The calibrated NCTUns simulator produced much more realistic outputs than the uncalibrated one. Then a signalized intersection was simulated in a case study to further investigate the packet drops of DSRC-based CV communications. The results indicated that there were significant packet drops, requiring further research before implementing safety critical CV applications.

Low-Cost and Energy-Saving Wireless Sensor Network for Real-Time Urban Mobility Monitoring System

This paper presents a low-cost and energy-saving urban mobility monitoring system based on wireless sensor networks (WSNs). The primary components of the proposed sensor unit are a Bluetooth sensor and a Zigbee transceiver. Within the WSN, the Bluetooth sensor captures the MAC addresses of Bluetooth units equipped in mobile devices and car navigation systems. The Zigbee transceiver transmits the collected MAC addresses to a data center without any major communications infrastructures (e.g., fiber optics and 3G/4G network). A total of seven prototype sensor units have been deployed on roadway segments in Newark, New Jersey, for a proof of concept (POC) test. The results of the POC test show that the performance of the proposed sensor unit appears promising, resulting in 2% of data drop rates and an improved Bluetooth capturing rate.

Evaluation of Variable Speed Limit under Connected Vehicle environment

The Variable Speed Limit (VSL) application provides travelers with dynamic speed advisory information to keep optimal traffic flow conditions for freeways and corridors under both recurrent and non-recurrent congestion incurred by incidents and/or work-zones. By coupling VSL and Connected Vehicle (CV) environment enabling two-way wireless communications for vehicle-to-vehicle and vehicle-to-infrastructure, it is expected to achieve improved performance of VSL as a viable traffic congestion mitigation tool. This digest paper presents the impact of connected vehicle on the effectiveness of VSL on a freeway bottleneck section by using a microscopic simulation model. Simulation experimental results show that CV-powered VSL improves the traffic congestion conditions up to 7-12% depending on CV market penetration rates.

Multiobjective Optimization Framework for Cooperative Adaptive Cruise Control Vehicles in the Automated Vehicle Platooning Environment

Automated longitudinal control technology has been tested through cooperative adaptive cruise control (CACC), which is envisioned to improve highway mobility drastically by forming a vehicle platoon with short headway while maintaining stable traffic flow under disturbances. Compared with previous research efforts with the pseudomultiobjective optimization process, this paper proposes an automated longitudinal control framework based on multiobjective optimization (MOOP) for CACC by taking into consideration four optimization objectives: mobility, safety, driver comfort, and fuel consumption. Of the target time headways that have been tested, the proposed CACC platoon control method achieved the best performance with 0.9- and 0.6-s target time headways. Compared with a non-optimization-based CACC, the MOOP CACC achieved 98%, 93%, 42%, and 33% objective value reductions of time headway deviation, unsafe condition, jitter, and instantaneous fuel consumption, respectively. In comparison with a single-objective-optimization-based approach, which optimized only one of the four proposed objectives, it was shown that the MOOP-based CACC maintained a good balance between all of the objective functions and achieved Pareto optimality for the entire platoon.

Multi-Thread Optimization for the Calibration of Microscopic Traffic Simulation Model

This paper proposes an innovative multi-thread stochastic optimization approach for the calibration of microscopic traffic simulation models. Combining Quasi-Monte Carlo (QMC) sampling and the Particle Swarm Optimization (PSO) algorithm, the proposed approach, namely the Quasi-Monte Carlo Particle Swarm (QPS) calibration method, is designed to boost the searching process without prejudice to the calibration accuracy. Given the search space constructed by the combinations of simulation parameters, the QMC sampling technique filters the searching space, followed by the multi-thread optimization through the PSO algorithm. A systematic framework for the implementation of the QPS QMC-initialized PSO method is developed and applied for a case study dealing with a large-scale simulation model covering a 6-mile stretch of Interstate Highway 66 (I-66) in Fairfax, Virginia. The case study results prove that the proposed QPS method outperforms other methods utilizing Genetic Algorithm and Latin Hypercube Sampling in achieving faster convergence to obtain an optimal calibration parameter set.

Smart Arrival Notification System for Americans with Disabilities Act Passenger Paratransit Service with a Consumer Mobile Device

This research presents an arrival notification system for paratransit passengers with disabilities. Almost all curb-to-curb paratransit services have a significantly large pickup time window, ranging from 20 to 40 min from the scheduled time and producing substantial passenger waiting times. The arrival notification system presented in this study delivers an automated voice call to a registered user once the paratransit vehicle is in proximity to the pickup location. The system utilizes the Google Traffic application programming interface (API) for the vehicle arrival estimation. Unlike other vehicle arrival notification systems in the state of the practice, the proposed system is compact and does not require additional equipment such as radio transmitting and positioning devices. The proposed system, which uses consumer mobile devices with the Android or iOS platform, is designed to exploit commercial cellular network service (i.e., 3G and 4G-LTE). In addition to the passenger notification, the proposed system provides paratransit drivers with real-time route guidance information developed through the Google Maps API. Field evaluation conducted in Essex County, New Jersey, revealed significant reduction in passenger waiting time. The passenger waiting time was reduced by 15 to 20 min. In addition, the accuracy of the notification system was tested. During the test, in almost all cases, the vehicle arrived 1 min earlier than the proposed arrival time.

Feature Representation and Extraction for Image Search and Video Retrieval

The ever-increasing popularity of intelligent image search and video retrieval warrants a comprehensive study of the major feature representation and extraction methods often applied in image search and video retrieval. Towards that end, this chapter reviews some representative feature representation and extraction approaches, such as the Spatial Pyramid Matching (SPM) , the soft assignment coding, the Fisher vector coding , the sparse coding and its variants, the Local Binary Pattern (LBP) , the Feature Local Binary Patterns (FLBP) , the Local Quaternary Patterns (LQP), the Feature Local Quaternary Patterns (FLQP) , the Scale-invariant feature transform (SIFT) , and the SIFT variants, which are broadly applied in intelligent image search and video retrieval .

Learning and Recognition Methods for Image Search and Video Retrieval

Effective learning and recognition methods play an important role in intelligent image search and video retrieval. This chapter therefore reviews some popular learning and recognition methods that are broadly applied for image search and video retrieval . First some popular deep learning methods are discussed, such as the feedforward deep neural networks , the deep autoencoders , the convolutional neural networks, and the Deep Boltzmann Machine (DBM) . Second, Support Vector Machine (SVM), which is one of the popular machine learning methods, is reviewed. In particular, the linear support vector machine, the soft-margin support vector machine, the non-linear support vector machine , the simplified support vector machine , the efficient Support Vector Machine (eSVM) , and the applications of SVM to image search and video retrieval are discussed. Finally, other popular kernel methods and new similarity measures are briefly reviewed.

Performance Evaluation of Video Analytics for Traffic Incident Detection and Vehicle Counts Collection

Current incident detection and traffic monitoring method using closed-circuit television (CCTV) cameras meets with limitations as the coverage of CCTV cameras rapidly expands. In general, traffic operators at Traffic Operation Center (TOC) have to manage and monitor numerous CCTV cameras deployed on roadways. Thus, many transportation agencies consider the use of video analytics system to reduce incident detection time and minimize traffic impacts, but they also want to validate the performance of the video analytics system whether it can work with their existing video surveillance infrastructure before procuring the system. To that end, a pilot study was designed and conducted to evaluate the accuracy of a video analytics product by integrating with CCTV cameras deployed on highways. The pilot study was designed to evaluate the accuracy of video analytics in detecting incidents and collecting traffic counts. The test results show that the performance of video analytics is significantly impacted by video quality and other environmental factors such as lighting and weather conditions.