Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
75 C h a p t e r 1 2 In this chapter, two conflict types that were mentioned in Chapter 5 are considered: left turn across path/opposite direc- tion (LTAP/OD), which is a legally possible maneuver at this intersection because of the presence of permissive left turns, and right turn into path (RTIP), which leads to the possibility of rear-end conflicts if the turning vehicle enters the path of a through vehicle traveling in the same direction. The approach is to apply triggers to screen for relevant motion types and impose a constraint on the times that the vehicles are closest to the intersection; if two vehicles follow the selected turn paths and are at the intersection within 10 s of each other, they represent a candidate conflict pair. For each case, additional constraints (e.g., for RTIP the turning vehicle should exit the intersection before the through vehicle) can be applied and the appropriate conflict metrics evaluated. In fact, for LTAP/OD, both timing cases where the left turning vehicle turns in front or behind are included. It is also possible to relate turning events to traffic signal state because these data exist in the database, but for this small-scale study (focused on demonstrating the usability of the data) no such analysis is performed. Left turn across path For LTAP/OD, postencroachment time (PET) is used as a metric (see Chapter 5). Recall that PET measures the time gap between the turning vehicle exiting the lane of the through vehicle and the time at which the through vehicle arrives. In Figure 12.1, the critical locations are shown with one vehicle trajectory shifted in time to the point of conflict (here for the turning vehicle ahead of the through vehicle); the size of the time shift equals the PET for the event. The extraction of these conflict events is completely automated, including vehi- cle positioning and size estimation; as described previously, length and width estimation is not fully mature in this analy- sis, so it is not all that surprising that the vehicle bounding boxes are somewhat distorted from reality. The actual vehi- cles are shown in Figure 12.2, in which the images shown are four frames apart (at 0.2-s intervals). The image is taken from the southwest camera, so we see the turning vehicle entering from the west and exiting to the north. All LTAP/OD conflicts were extracted. However, very few events actually took place. For the full data set, only 67 permis- sive left turns were taken, with most (47 cases) with the turn- ing vehicle entering from the east and exiting to the south. Figure 11.1 shows that permissive left-turn events (from the east and west) are relatively rare. Of these events, only 15 cases included a conflict of the type shown above, with the turning vehicle generating an LTAP/OD conflict with PET <10 s. If other (mild) conflicts with negative PET values (where the turning vehicle waits for the oncoming vehicle to pass before turning) are included, another 47 events were found. Distributions are shown in Figure 12.3. Even with small data volumes it is clear that time separation is much less in the second case. right turn into path As a second example of conflict metrics, RTIP events are con- sidered. Starting with the full set of trajectories, all right turns are found, and for each case corresponding straight paths are sought. All pairs of trajectories meeting these requirements and with times nearest the intersection center within ±10 s are analyzed further. Table 12.1 shows that 2,764 such candi- date events were found; in most cases, the turning vehicle arrives from the west and turns south and the through vehicle arrives from the north. Of 1,892 cases, in only 11 do vehicles exit to the east. For these events three successive filters are applied: (1) the turning vehicle must exit the intersection first, (2) the through vehicle should the first in any group paired with the turn- ing vehicle, and (3) the turning vehicle should be last of any Conflict Analysis
76 Figure 12.1. Conflict point for left turn across path (PET îµ 1.78 s). Figure 12.3. Distribution of postencroachment times for permissive left turns (62 cases). 0 1 2 3 4 5 6 7 8 9 10 0 0.5 1 1.5 2 2.5 3 PET (sec) Fr eq ue nc y Turning Vehicle Crosses First -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 0 2 4 6 8 10 PET (sec) Fr eq ue nc y Turning Vehicle Crosses Second Figure 12.2. Path-crossing conflict (LTAP, PET îµ 1.78 s). Pi xe ls Pi xe ls Pi xe ls Pi xe ls 0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300 0 50 100 150 200 250 300
77 their positioning is not the most accurate available. Note that there is only one clearly corrupt track in the set, and this is screened out in the next stage, which is to apply the KF to the 367 cases and use speeds and distances to calculate TTC time histories. If the minimum TTC is less than 10 s, the event is retained. Figure 12.5 shows a typical result. In this event, both vehi- cles exit to the north; distance is measured in the northerly direction to calculate the valid TTC, even when the turning vehicle is not yet moving due north; the black lines, repre- senting the turning vehicle, are cut off at the beginning so values are found only for when both vehicles are in the through lane. In this case, tracking of the through vehicle started a little after the turning vehicle entered its path, so its position was extrapolated assuming constant speed; the extrapolation is shown in the thicker line. Figure 12.6 shows the time history of TTC during the event, for which the minimum value is the defined conflict metric. Table 12.1 shows there are 200 such events. Interestingly, the through vehicle is seen to stop (Figure 12.5). This is not because of hard braking to avoid collision (in fact turning vehicles paired with the lead through vehicle. These filters drastically reduce the number of cases, to 367, as seen in Table 12.1; however, the proportions of cases are roughly maintained. The plots in Figure 12.4 show trajectories of the 367 cases, where red indicates the through vehicle and black dashed lines indicate the turning vehicle. These are simple cluster tracks used for approximate timing and screening, so Table 12.1. Counts of RTIP Events as Successive Filters Are Applied Turning Vehicle Direction WS NW EN SE TotalThrough Vehicle Direction NS EW SN WE Coincide within 10 s 1,832 38 823 11 2,764 With turning vehicle leading 416 4 162 2 584 Filtered for nearest vehicle in group 253 4 109 1 367 TTC <10 s 135 3 62 0 200 Figure 12.4. Vehicle trajectories used for TTC analysis. -60 -40 -20 0 20 40 60 80 -80 -60 -40 -20 0 20 40 Meters M et er s -50 0 50 -40 -30 -20 -10 0 10 20 30 40 50 60 M et er s Meters -60 -40 -20 0 20 40 60 -40 -20 0 20 40 60 M et er s Meters -40 -20 0 20 40 60 -50 -40 -30 -20 -10 0 10 20 30 40 M et er s Meters
78 was 1.4 s. There is a clear pattern of conflict frequencies increas- ing roughly linearly with increasing TTC value. The ability to draw conclusions or make comparisons between different turn- ing directions, different signal states, traffic densities, weather conditions, and so forth is impossible with such a small event set. However, such analysis would be feasible if more extensive data collection takes place in the future. no such events took place in the field test). However, check- ing the traffic signal state and reviewing recorded video, this vehicle slows to a near-stop for a red traffic signal, then the signal turns green and the vehicle accelerates. The turning vehicle simply took advantage of this opportunity. Finally, Figure 12.7 shows the results of the remaining 200 events. No event had TTC less than 1 s, and the minimum time 374.5 375 375.5 376 376.5 377 5 10 15 time (s) TT C (s) Figure 12.6. TTC time history during the conflict event of Figure 12.5 (TTCmin îµ 5.6 s). Figure 12.5. RTIP conflict. Red: through vehicle; black: turning vehicle; blue circle: point of minimum time to collision. 374 376 378 380 382 384 386 388 -100 -50 0 50 100 time (s) di st an ce (m ) 374 376 378 380 382 384 386 388 -5 0 5 10 15 time (s) sp ee d (m /s) Figure 12.7. Distribution of time-to-collision times, right turn into path (200 events). 0 1 2 3 4 5 6 7 8 9 10 0 5 10 15 20 25 TTC (sec) Fr eq ue nc y Right Turn into Path
