Evaluating Location of Placement and Spacing of Detectors for Real-Time Crash Prediction on Urban Expressways

The concept of real-time crash prediction is in its early stage and so far the focus has been on evaluating different methods to improve the overall prediction accuracy assuming a fixed existing underground loop detector infrastructure. Very few studies are conducted on the location of placement and spacing of detectors for urban expressways which is vital when the crash potential of a specific road section is to be monitored. Guidelines regarding this can be highly beneficiary for developing proactive road safety management systems under budget constrains for hazardous road sections. This study evaluated six different detector spacings by developing six separate real-time crash prediction models with Bayesian Network. Crash data and traffic flow data (flow and speed) were collected for two years (December, 2006 to November, 2008) from 30 loop detectors on Shinjuku 4 Tokyo Metropolitan Expressway, Japan. Crash data were aggregated for each 250 meters road segment and six different detector spacings (in each case, one detector in the upstream and the other one in the downstream of the segment under consideration) were evaluated. The 5-minute cumulative flow difference and average speed difference between upstream and down stream detectors were found to be the suitable predictors. Although all the six detector spacings could predict more than 50% future crashes accurately, the spacing with the downstream detector placed 500 meters and upstream detector placed 250 meters from the center of the 250 meter road segment provided 63% accuracy for crash and 80% accuracy for non-crash situations.