Real-Time Traffic Queue Length Estimation at the Freeway Off-ramp Using Dual-Zone Detectors

Congestion at the downstream of a freeway off-ramp often propagates the traffic queue to the mainline, and thus reduces the freeway capacity at the interchange area. Hence, an accurate queue estimation model can help the traffic engineers to select proper control strategy to mitigate queue spillback at off-ramps. In responds to such a need, this study presents a queue estimation system using dual-zone detectors, where each detector can provide two detection zones. Particularly, the short detection zones are used to count traffic flow rates and the information from long detection zone can indicate the presence of traffic queues. With respect to different congestion levels, the off-ramp queue may be fully discharged during the green time or not. Therefore, this study has developed two queue estimation models for each condition. Based on the field data from the freeway interchange in Zhubei, Taiwan, this study has conducted extensive simulation experiments, and demonstrated the effectiveness of the proposed system on off-ramp queue estimations. INTRODUCTION Since most drivers do not tend to segregate themselves by destination well in advance of an off-ramp, but rather make most of their lane-changing decisions at the last moment. The exit queue of an off-ramp might spread itself laterally upstream of an off-ramp, thereby restricting the efficiency of the mainline flows. Hence, congested conditions at downstream intersections can lead to a long traffic queue at the off-ramp, and the queue spillback may propagate to its upstream and block freeway travel lanes. To mitigate freeway congestion caused by the excessive off-ramp queue, a reliable and accurate queue estimation model is essential for off-ramp controls. In practice, two types of detectors are widely used to capture traffic information: video sensors and loop detectors. Through image interpretations and digital analysis, video sensors can verify the arriving vehicles and consequently capture the queue information. Due to some impact factors such as weather condition, visibility obstacles, and the reliability of image processing, the obtained queue information may not accurately reflect the real traffic conditions. Video sensors are usually installed near intersections for queue identifications. However, for off-ramp queue detections, video sensors may be incapable to cover the entire ramp due to their limited detection scope. In contrast, loop detectors are more reliable compared with video sensors. In review of the literature, practical approaches usually use one or more loop detectors at the upstream or downstream of links for queue estimations. Following the similar principle, the objective of this study is to develop a reliable queue estimation model based on loop detector measurements. The development of technology has promoted various types of wireless loop detectors, such as Radar, Microwave and Infrared detectors. Dual-loop radar detector is possibly one of the most widely used detectors in the dynamic traffic control systems. Within the pre-defined detection zones, the detector is able to verify the presence of arriving vehicles and provide the flow rate and vehicle speed. Based on the experience of our field implementations, the presence data is the most accurate information provided by radar detectors. Hence, the queue estimation model developed in this study will only use the presence data for calculations. Also, the length of detection zone (distance between two loops) can affect the function of detectors. For instance, short detection zone can be more efficiently to count the number of vehicles while observing 100% occupancy measurements can indicate stopped vehicle over the long detection zone. To take advantage of the both functions, a dual-zone detector is used in this study for providing both short and long detection zones. This paper is organized as follows: in the next section, a literature review for existing queue estimation techniques at signalized links is provided; then the implementation of detectors along with the presence data analysis are provided in the following section; in response to different traffic conditions, section 4 presents two queue estimation models; using a network in Zhubei, Taiwan for our study site, the proposed models are tested in section 5; key findings and conclusions are summarized in section 6. LITERATURE REVIEWS Queue estimation is quite critical for signal optimization since it is one of the most crucial performance measurements in an intersection (Newell, 1965, Webster, 1958, Balke et al., 2005, Mirchandani and Zou, 2007, Lu and Yang, 2014). A lot of researches focus on average or maximum queue length estimation in a signal cycle or during a period of time. Loop detectors are often used when researching queue length estimation and various methods have been reported. The input-output method counts the accumulative arrival at the rear of the queue and accumulative discharge at the front (Sharma et al., 2007; Vigos et al., 2008). However, the major limitation of the input-output method is basically caused by the fact that it cannot handle long queues exceeding the rear detector. Skabardonis and Geroliminis (2008) estimated intersection queue length with aggregated 30-sec loop detector data based on the shockwave theory. Their methods examine the flow and occupancy data every 30 seconds to estimate the rear end of the queue. It can theoretically handle a long queue length even if it exceeds the location of the rear detector but the 30-sec aggregation smoothies the variation the traffic pattern and makes the method less sensitive to the change of traffic state. Similarly, Smaglik et al., (2007) and Liu et al. (2009) also uses event-based data, which means that the traffic state change can be identified by investigating the real-time data and therefore queue length can be estimated. Vehicle trajectory data was made possible by probe vehicle technology. Cheng et al., (2011) proposed a method based on sampled vehicle trajectories as the only input. The concept of critical point is introduced to represent the changing vehicle dynamics. This method is also one of the shockwave methods and is evaluated by a recently collected data set from a GPS logger. Mobile sensors are also used to estimate real-time queue length. Ban et al., (2011) examines the discontinuities and non-smoothness in travel time data from mobile sensors, which indicates signal timing or queue length changes. The concept of Queue Rear No-delay Arrival Time is then introduced after the maximum and minimum queue length in a cycle has been estimated. Despite the research progress on real-time queue estimation at signalized intersections, the unique characteristic of off-ramp queue may require a new model for estimation. For instance, the long distance between upstream and downstream of off-ramp may cause extremely long queues. Also, the occurrence of spillover at downstream link can directly impact the queue discharging process. By investigating the relations between time occupancy and indirectly measureable density, Qian et.al (2012) developed a queue estimation model over signalized off-ramps. Their numerical example demonstrates satisfactory estimation accuracy in the simulation tests. However, the proposed method has shown relatively unsuccessful results in capturing short queues. Also, only total number of vehicles within the off-ramp is estimated in this study. DATA DETECTION AND ANALYSIS As aforementioned, the distance between two loops will determine the function of detectors. To take advantage of both short and long loop detectors, dual-zone detectors are implemented at both upstream and downstream of off-ramp, as shown in Figure 1. Figure 1 Location of dual-zone detectors on the target off-ramp Since the presence data is much more accurate and reliable than the other data provided by the detectors, this study only use presence data for calculations. In practice, a “0-1” format data with short interval (e.g. 0.1 second) may be obtained from the detectors, as shown in Figure 2. For the convenience of discussion, the following analysis and calculation will assume a “0-1” format data are available. Upstream Detector Downstream Detector Short Detection Zone Long Detection Zone Figure 2. An illustrative example for the presence data format Based on the obtained data, the emerging of multiple continuously “1” or “0” can indicate the traffic conditions over the detection area. As shown in Figure 3(A), the presence of “0” from short detection zone can be used to record the number of passing vehicles within the target time period. Similarly, for long detection zones, multiple “1” can reflect the formation of queue and the presence of “0” will indicate the clearance of queue. Time 0 0 0 1 1 1 0 0 0 1 1 1 1 0 0 0 1 1 1 One Vehicle One Vehicle One Vehicle Figure 3(A) The recording of number of passing vehicles by the short detection zone Time 0 0 0 1 1 1 1 1 1 1 1 1 1 0 0 0 0 1 1 Formation of queue Queue Clearance Figure 3(B) The detection of queue formation and clearance by the long detection zone Obviously, some detection errors may exist in practice using the deification methods introduced above. As shown in Figure 4(A), when using the short detection zone to record number of passing vehicles, if the loop distance is longer than the headway between two adjacent vehicles, it may be identified as one large size vehicle. Hence, to ensure the estimation accuracy, the loop distance of short detection zone should be shorter than the minimum vehicle headway. Figure 4(B) shows the detection errors under two possible conditions. If the loop distance is short than the stopping vehicle headway, the detector may not occasionally identify the formation of queue. Also, when the loop distance is longer than the headway between two moving vehicles, the model may mistakenly identify a queue. 0 0 0 1 1 1 1 1 1 1 0 0 0 0 One Largesize Vehicle Figure 4(A) Identification errors caused by short detection zone Time 0 0 0 1 1 1 0 0 0 1 1 1 1 0 0 0 1 1 1 Vehicle Presence Vehicle Presence Vehicle Presence