Journal of Intelligent Transportation Systems最新文献

Classification of driving events is a crucial stage in driving behavior monitoring using smartphone sensory data. It has not been previously explored that to what extent classification performance depends on the classifier type and input data characteristics. To fill this gap, a real-world experiment is designed for supervised data collection. Then the effects of different machine learning (ML) classifiers, data sampling rates, and sensor combinations on the final classification accuracy are demonstrated. A considerable number of labeled events (4114) containing 11 types of driving maneuvers are collected using base sensors (accelerometer and gyroscope) and composite sensors (linear accelerometer and rotation vector) available in smartphones. Several models using 23 ML algorithms are trained. The sensitivity of these models is analyzed by changing the characteristics of the input data concerning the type of ML classifier, data sampling rate, and the bundle of mobile sensors. It is demonstrated that: (1) F1 scores vary from 70 to 96% for different ML classifiers, (2) F1 scores drop 30–40% depending on the classifier type when reducing the data sampling rate, and (3) using all four sensors as a bundle for classifying driving events is not reasonable since an approximate equal F1 score is achievable by a three-sensor bundle which includes an accelerometer and a linear accelerometer.

This study develops a data-driven approach for identifying critical transfer zones in the city to facilitate the coordination of transit and emerging on-demand services. First, the methods convert the trajectories into a 3 D grid with an optimal cube size. Built upon that, we zoom in and study the trajectory density of each mode in a cube and present the results by heatmaps. After that, we zoom out and aggregate those cube information fragments through the clustering algorithms to explore two critical patterns: the ridesharing swarm (RS) zones where many ridesharing trips go through, and the “sandwich pattern” zones where a transit trajectory dominant zone is sandwiched by two ridesharing trajectory dominant zones. Our numerical analysis confirms that these RS zones are well correlated to the promising areas/corridors for integrating transit and on-demand services; the “sandwich patterns” help discover first/last mile (FLM) zones. Last, we further develop a two-channel deep learning network to predict the variation of the FLM gaps so that adaptive services can be planned. A case study based on the field data of the second ring region of Chengdu, China confirms the effectiveness and capability of our analysis approach.

The electric-scooter (e-scooter) has become a popular mode of transportation with the proliferation of shared mobility services. As with other shared mobility services, the operation of the e-scooter sharing service has a recurring problem of imbalance in supply and demand. Various strategies have been studied to resolve the imbalance problems, including demand prediction and relocation strategies. However, the difficulty of accurately predicting the fluctuating demand and the excessive cost-labor consumption of relocation are major limitations of these strategies. As a remedy, we propose a deep reinforcement learning algorithm that suggests price incentives and an alternative rental location for users who find it difficult to acquire e-scooters at their desired boarding locations. A proximal policy optimization algorithm considering temporal dependencies is applied to develop a reinforcement learning agent that allocates the given initial budget to provide price incentives in a cost-efficient manner. We allow the proposed algorithm to re-use a portion of the operating profit as price incentives, which brings higher efficiency compared to the same initial budget. Our proposed algorithm is capable of reducing as much as 56% of the unmet demands by efficiently distributing price incentives. The result of the geographical analysis shows that the proposed algorithm can provide benefits to both users and service providers by promoting the use of idle e-scooters with a price incentive. Through experimental analysis, optimal budget, i.e., the most efficient initial budget, is suggested, which can contribute to e-scooter operators developing efficient e-scooter sharing services.

Principal objectives of the Intelligent Transportation Systems (ITS) are to improve traffic safety, facilitate informed traffic decision making, and enhance quality of life and services in a smart traffic environment. Vehicle crashes at urban traffic intersections are among the rudimentary sources of injuries and fatalities in the cities. According to the report of the World Health Organization (WHO), in every 25 seconds, one vulnerable road-user is being killed by a vehicle crash. Therefore, it is necessary to take a novel and smart approach for improving the safety and reducing vehicle crashes. This leads to a contextual perception and spatial awareness of driver to increase security and safety for the driver, vehicle, and road users. Autonomous vehicles collects the information from the environment through equipped sensors on the vehicle such as camera, laser, radar, and Global Navigation Satellite Systems (GNSS). The main challenge arises when the person or objects are located beyond the driver’s Field of View (FOV) and cannot be detected by embedded sensors on the vehicles. This paper proposes an Advanced Driver Assistance System (ADAS) to increase the safety on road intersections by taking advantage of existing infrastructures (e.g road camera) being used for traffic control. The aim of this research is improving the driver’s FOV using a computer vision approach (e.g background subtraction algorithm) and Location Based Service (LBS). The case study results at Tehran metropolitan demonstrate the reduction in traffic collision risk and improvement of pedestrian safety using the proposed system.

Flight time estimation is expected to play a crucial role in predicting the Estimated Time of Arrival, which could help detect conflicts and manage arrivals. This paper proposes a novel data-driven method for flight time estimation based on arrival pattern identification and prediction. Firstly, a trajectory clustering algorithm is employed to group the arrival trajectories into different arrival patterns. A new trajectory representation technique is presented during the clustering process for better-describing arrival patterns. Secondly, we extract features from radar tracks for data-driven flight time estimation. These features consist of current states related, historical information related, traffic situation related, and environmental conditions related features. Furthermore, the permutation feature importance and recursive feature elimination method are adopted to reduce feature dimensions. Then, we develop three widely used tree-based models to estimate the flight time for each arrival pattern. We also propose an image-based flight patterns prediction method to classify each new arrival aircraft into the corresponding arrival pattern for actual operation. Finally, we take the Guangzhou arrival operation as a case to validate our proposed method. The results indicate that our proposed method could improve flight time estimating accuracy. Besides, through the data-driven strategy, we could also find several significant factors affecting the flight time within the Terminal Maneuvering Area.

Following the most energy-efficient route can have a significant impact on reducing energy consumption. While most eco-routing research has focused on reducing energy consumption and travel time, the safety aspect of route choice is currently neglected. In this paper, a multi-objective optimization methodology is formulated to concurrently minimize fuel consumption, travel time and safety risk, which is quantified using a novel methodology based on network-level safety measures. The proposed optimization framework provides a transparent way to intuitively include driver preferences via “budgets” for time, fuel consumption and safety – which represent the driver’s willingness to sacrifice these factors for fuel consumption improvements. The performance of the proposed method was tested on urban road networks in Brisbane-Australia, with a rear-end collision risk model as the safety measure. The results demonstrate that eco-routing with safety considerations has the potential to improve fuel efficiency while simultaneously reducing safety risks.

As a special application of connected and automated vehicles (CAVs), the Autonomous Truck Mounted Attenuator (ATMA) vehicle system is promoted to reduce fatalities in work zone locations. In this manuscript, we focus on the Operational Design Domain (ODD) problem of two-lane highways, i.e., under what traffic conditions should an ATMA be deployed. Due to the dramatic speed difference between ATMA vehicles and general vehicles, a queue will be formed, leading to a percent-time-spent-following (PTSF) increase during maintenance. General vehicles in the queue will assess a gap on the opposite lane to perform a passing maneuver, which is broken down into multi-stage merging behavior. As such, an analytical model is first made, based on queuing theory in which the arrival rate and service rate are analyzed to estimate the PTSF. In this way, the linkage between annual average daily traffic (AADT) and level of service (LOS) is analytically established. Then, the proposed model is validated by comparing the estimated PTSF with that of the Highway Capacity Manual (HCM) values. The comparison results show that the mean error is 9.58%, and the mean absolute error is 12.36%, which demonstrate that the developed model is able to generate satisfactory results when compared with the HCM model. Numeric analysis also shows that roadway performance is sensitive to the K factor and D factor, as well as the operating speed of an ATMA. If LOS = C is a desirable design objective, a good AADT threshold to use would be around 11,000 vehicles per day.