When looking at measures of fatal and severe-injury crashes, roundabouts have demonstrated improved safety performance compared to traditional signalised intersections. This, coupled with their delay-reduction potential, has made roundabouts attractive to practitioners and has helped drive their adoption throughout the United States. Despite this, when it comes to less severe crashes, multilane roundabouts fail to provide the same benefit. In many cases, they actually come with increased rates of crashes resulting in property damage only. The prevalence of driver errors and property damage crashes at “2x2” roundabout conflict locations, where two entering through lanes cross two exiting through lanes, may deter agencies from implementing such designs in favour of single-lane roundabouts, which provide less mobility and shorter design life, or traffic signals, which have higher delays and are less safe. The effects of this can be seen across Minnesota, where several 2x2 roundabouts have recently been converted to 2x1 configurations by changing striping to reduce the number of available through lanes. At best, this is only a temporary solution, however, since the additional capacity will likely be needed in less than 20 years. In the long term, a more effective solution to this problem requires determining why drivers are prone to certain mistakes and developing a design that can correct this behaviour. Previous research into this topic has identified the following behaviours that are associated with crashes in two-lane roundabouts, with yielding violations and turn violations generally being the largest contributors to crashes: *Yielding violations (entering vehicles failing to yield to cross traffic); * Lane changes inside the roundabout; * Turn violations (e.g., right turns from the left-hand lane); and * Wrong-way movements A previous observational before/after study conducted by the University of Minnesota focusing on a 2x2 roundabout in Richfield, MN, has demonstrated that changes to signs and lane markings of the roundabout significantly reduced the occurrence of some of these errors (Richfield et al., 2013). While that roundabout is atypical both in terms of design (pre-2009 MUTCD) and driver population (young suburban and commuter traffic), the results of the study are still encouraging, as they suggest that effective solutions for treating these common mistakes can be found as more data is collected and the problem becomes more well understood. This study seeks to build on this work by expanding the data collection effort to include more sites, collecting observations of undesirable driving manoeuvres at several of the remaining 2x2 roundabouts in Minnesota and relating the frequency of individual behaviours to specific design features. Ultimately, four roundabouts throughout the state were chosen for data collection and analysis, with two of being them fully multi-lane roundabouts with exclusively 2x2 conflict areas and the other two being partial multilane roundabouts including both 2x1 and 2x2 conflict areas. These locations are: University Drive South and 5th Avenue South in St. Cloud (partially 2x2); 185th Street West and Kenwood Trail in Lakeville (fully 2x2); TH-22 and Adams Street in Mankato (partially 2x2); and TH-22 and Madison Avenue in Mankato (fully 2x2) These roundabouts have all been built recently and are based on the latest design standards, which prescribe much larger deflection angles for entrances and exits compared to Richfield, but serve very different driver types and exhibit enough variation between roundabouts to help highlight which design features are most effective at reducing a given type of violation. This study largely follows the same experimental design successfully executed in the previous study with some updates to automate parts of the process as well as to extract more data for analysis. This methodology involves the collection of several days’ (or more) of video observations and the identification of all the driving violations performed by the drivers. Whereas the earlier study relied on manual reduction of the video records, this process was automated for the current study by using the computer vision application, TrafficIntelligence, to extract vehicle trajectories directly from video then using a custom trajectory analysis to tabulate data for analysis. This not only reduced the time required to process the video, but it also allowed turning movements and speeds to be output for every vehicle observed using the roundabouts. This provided increased ability to control for the effect of traffic in the roundabout on the rates of undesirable behaviours and allowing the underlying relationship with design features to be seen. The analysis of this data focused on yielding and turn violations, which are the most commonly, cited reasons for crashes occurring in two-lane roundabouts. Once violation and volume data were extracted, the data was analysed to determine how the rates of these violations vary by location and relevant design features. In regards to Turn Violations, the roundabout in St. Cloud presents some differences when compared with the roundabout in Richfield. Specifically, left-from-outer-lane violations, which are the source of the most serious crashes, exhibit noticeably lower rates. Several possible causal factors for this were explored with no revealed correlation. The one geometrical difference in this roundabout is that all roads approaching the roundabout have one lane per direction upstream of the roundabout entrances while, in Richfield, all roads had two lanes per direction upstream of the roundabout. One can hypothesize that, in the case of left turns, drivers instinctively choose to stay close to the left curb and, by extent, the inner lane of the approach. In the Lakeville roundabout, the left-from-outer-lane violations are similar to Richfield, except on the approaches where the upstream direction has one lane per direction, reinforcing the aforementioned hypothesis. Both of the Mankato roundabouts have overhead lane designation signs, which seems to be directly related to much lower turn violations observed as compared to Richfield and the other roundabouts in this study. This study has put extra effort into understanding the causes of yield violations since the earlier study failed to produce a traffic control plan that can reduce their rate of occurrence. Unfortunately, this study, too, did not produce insight on the nature of the problem or potential solutions. In the St. Cloud and Lakeville roundabouts, the rates of yield violations followed more or less the same rates as in the earlier study and followed the familiar pattern of higher rates in failures to yield to the inner lane of the roundabout. The failure-to-yield (FTY) rates in the in the approaches from Adams Street also occurred at similar or slightly higher levels. However, the TH-22 NB approach presented significantly lower rates, almost 10 times lower, while the TH-22 SB approach was somewhere in between. The latter has only one lane of cross traffic to yield to. The results are a little more complicated at the Madison Ave Roundabout. Specifically, the approach from Madison EB exhibited an initial rate that was slightly lower than the norm compared to the earlier study and was sharply reduced after the learning period was over. Unfortunately, in the period after the traffic control changes, it climbed again to nearly the same level as during the learning period though that rate was still lower than the norm. There is no real explanation for this other than the traffic control changes not having the desired effect. The approach from TH-22 SB follows the same pattern as the Adams Roundabout, with the FTY rate being significantly lower than any other roundabout in the study and remaining unchanged throughout the three study periods. The conclusion is that the traffic control changes implemented at the Mankato roundabouts did not produce any significant improvements on the yield violation problem. (Author/publisher)
Abstract