This study examines the association between road geometry, land uses, access points, and streetscape environments with pedestrian-vehicle crashes (2018-2020) on urban arterials in Austin, Texas, using negative binomial models. This study assessed streetscape environments using a computer vision approach driven by machine learning algorithms and Google Street View (GSV) images. The results showed that arterials with high posted speed limits were associated with increased numbers of total, fatal, and injurious crashes. Traffic-generating uses (i.e., commercial and office uses) were associated with increased pedestrian-vehicle collisions. Arterials abundant in green spaces were linked to a reduced number of total and fatal crashes.