Performance measures and input uncertainty for pedestrian crossing exposure estimates.

Pedestrian safety performance measures often use estimates of annual crossing exposure as inputs-but relatively little information exists on the uncertainty associated with these inputs. This research considers two sources of temporal information for expanding short-term counts: (1) a composite of pedestrian counts from other cities, and (2) local vehicle counts. A database of pedestrian flows from video review covering 12 months and including over 350,000 pedestrian observations provides a known reference annual volume and a set of short-term counts for expansion and testing. The research compares the temporal information sources with observed pedestrian volumes by analyzing the times and magnitudes of volume peaks. The temporal patterns based on local vehicle counts match observed pedestrian patterns more closely than the external composite pedestrian patterns. To quantify exposure estimate uncertainty, the research uses the local vehicle and external composite pedestrian patterns to expand a sample of short term counts to generate a set of 200 annual estimates, and then compares the estimates to the known reference volume. Exposure estimates developed by expanding counts with local vehicle factors have the lowest errors (mean: -2%; median: -3%, standard deviation: 33%; 90 percent of errors between -53% and 50%). Exposure estimates based on external composite pedestrian patterns have higher errors (mean: 27%; median: 9%; standard deviation: 73%; 90 percent of errors between -62% and 170%). If methods to obtain pedestrian exposure estimates based on short-term counts are improved, more confidence can be placed in safety performance measures that use these estimates as inputs.