Explainable, automated urban interventions to improve pedestrian and vehicle safety

Abstract

At the moment, urban mobility research and governmental initiatives are mostly focused on motor-related issues, e.g. the problems of congestion and pollution. And yet, we can not disregard the most vulnerable elements in the urban landscape: pedestrians, exposed to higher risks than other road users. Indeed, safe, accessible, and sustainable transport systems in cities are a core target of the UN's 2030 Agenda. Thus, there is an opportunity to apply advanced computational tools to the problem of traffic safety, in regards especially to pedestrians, who have been often overlooked in the past. This paper combines public data sources, large-scale street imagery and computer vision techniques to approach pedestrian and vehicle safety with an automated, relatively simple, and universally-applicable data-processing scheme. The steps involved in this pipeline include the adaptation and training of a Residual Convolutional Neural Network to determine a hazard index for each given urban scene, as well as an interpretability analysis based on image segmentation and class activation mapping on those same images. Combined, the outcome of this computational approach is a fine-grained map of hazard levels across a city, and an heuristic to identify interventions that might simultaneously improve pedestrian and vehicle safety. The proposed frame

Predicting the impact of urban change in pedestrian and road safety

“…Further, simple and well-organized scenes seem to be related to lower accident rates. The causal plausibility of this uncovered correlation can be supported by extensive literature on attention processes 15,[33][34][35][36] , which confirm that complex scenes are harder to process (in terms of division of attention) by drivers and pedestrians, which could logically lead to an increase in driver's stress 37 or accident rates 10 .…”

“…With vast amounts of data about cities currently available, automated analysis has been recognized as a crucial tool to help urban planners in decision tasks [3][4][5][6] . While the link between the growth of urban open data and new computational techniques such as Deep Learning has been pointed out by several authors [7][8][9][10][11][12][13] , much work can still be done in leveraging these new technologies towards the end of achieving the ambitious goals of Vision Zero.…”

“…Up to now, due to lack of available (or open) data and tools for automated analysis, these studies have mostly been constrained either to the limited features available from planimetric map data 16,17,23 , or to detailed, but time-consuming, descriptions of particular settings collected by hand 20,21 . However, the rise of new Big Data sources 13,[24][25][26] related to urban environments and transportation has boosted the development of new techniques 10,[26][27][28] and the combination of complementary tools existing in different fields 5,29 , such as complex systems 18,30 .…”

“…In previous work 10 , with the aim of clarifying the relation between pedestrian and vehicle accidents and built environment features, we compiled and analyzed a large dataset of street-level images obtained from Google Street View 31 and the crowdsourced Mapillary 32 collection. These datasets provide valuable fine-grained information on the location and organization of urban objects, and even dynamic information in terms of existing mobile objects (pedestrians, bicycles, cars and public transport).…”

“…These datasets provide valuable fine-grained information on the location and organization of urban objects, and even dynamic information in terms of existing mobile objects (pedestrians, bicycles, cars and public transport). The Deep Learning model in 10 , taking as inputs (1) driver point-of-view images, and (2) geotagged historical accident data, shows notable performance in predicting accident rates at a very high spatial resolution (approx. 15 meters).…”