Evaluation of pedestrian critical gap and crossing speed at midblock crossing using image processing.

Pedestrians confront risky situations at midblock sections due to the unyielding behavior of drivers. Thus, pedestrians have to wait for an appropriate gap to cross. This research investigates pedestrians' gap acceptance and crossing speed for midblock crossings by image processing methods in Izmir, Turkey. A total of 498 pedestrians have been tracked at two midblock crossings. The data were collected for one hour at each midblock crossing during the evening peak hour between 5.00-6.00 p.m. Three synchronized cameras were used to record pedestrian crossings. Then, by using image processing, vehicle and pedestrian trajectories have been obtained. Two cameras were mounted on telescopic tripods reaching up to 9 m, and the third camera was used to identify pedestrians' gender better. The parameters extracted from the recordings are; pedestrians' gender, group size, whether they carried items or not, and their accepted/rejected gaps. Pedestrian and item detection has been performed by YOLOv3 and YOLACT models. The accepted and rejected time gaps were extracted for pedestrians, excluding the pedestrians who crossed between stopped vehicles and crossed when an approaching vehicle did not exist within 100 m from the midblock crossing. Raff's method was used to estimate the critical gap using accepted/rejected gaps. The critical gaps ranged between 4.1 s and 6.2 s. The 15th percentile crossing speeds were found to be similar, ranging between 0.78 m/s and 0.80 m/s.