This paper models road traffic collision counts recorded between 2015 and 2019 in a ward located in the central part of Cape Town in South Africa, using a Bayesian spatio-temporal zero-inflated Negative Binomial approach. The method accounted for the excess zeros present in collision data by separately modeling zero and non-zero collision counts, while also capturing spatial and temporal dependencies through prior distributions. Road-level information was used as fixed-effects covariates, including speed limits, presence of traffic calming measures, traffic signals, road class, number of lanes, whether the intersection is on “Main Road”, and whether a public transport route passes through the intersection. The results reveal that among the covariates included in the selected model, node degree (used as a proxy for traffic flow), the presence of traffic signals, having any major road around the intersection (road class), location along “Main Road”, and the presence of a taxi route at the intersection were all associated with an increase in traffic collision counts at the intersections. The years 2018 and 2019 were associated with higher collision counts compared to the reference year, 2015. For the probability component of the model, the existence of traffic signals at the intersection and location along “Main Road”were both associated with an increase in the chances of at least one collision being observed at the intersection, whereas having any high-speed road around the intersection decreased this chance.