A framework to evaluate the performance of bias correction to vehicle counts

2 This paper proposes a new framework to evaluate the performance of correcting systematic biases 3 commonly latent in vehicle counts measured at freeway loop detectors. Through the proposed framework, 4 one can evaluate the performance of bias-correction by evaluating the legitimacy of density estimates 5 generated with bias-corrected counts. To test this framework, traffic data for a 1.2km-long freeway site 6 (on Interstate 5 Northbound in Sacramento) were collected over 30 weekdays both from its loop detectors 7 and from probe vehicles traversing it. These heterogeneous traffic data in combination were processed to 8 construct speed-density as well as speed-occupancy plots for individual freeway segments constituting the 9 study site. The test outcome turns out quite promising if we consider that probe data used for this study 10 show low average penetration rates, amounting to only three or four vehicle trajectories per hour. A 11 secondary purpose of this study is to develop and test a heuristic method to correct count biases based on 12 the conservation-of-vehicles principle. This bias-correction approach is designed to distribute total count-13 biases, accumulated at erroneous detectors, over all the intervening 30-second time intervals in proportion 14 to the counts newly added at the problematic detectors during each time interval. This approach can be 15 refined by grouping time intervals as needed to make the proportional bias-correction formula tailored for 16 each time-interval group. This refined bias-correction turns out to perform properly in most cases. 17 18 Keywords: loop detector errors, vehicle count biases, bias correction, density estimation, speed-density 19 plot 20 21 2 1