Accurate detection and localization of objects are key aspects of connected mobility and play pivotal roles in ensuring road safety. In particular, precise localizations are crucial for predicting potential collisions between vehicles and vulnerable road users (VRUs) crossing streets. This paper presents an evaluation of the accuracy of our camera-based roadside infrastructure in the context of collision risk detection. Specifically, our evaluation entails the deployment of two differently oriented cameras capturing a scene where a vehicle and a pedestrian converge towards a common point. By comparing camera-acquired object positions with ground truth data obtained through GNSS RTK-equipped objects, we evaluate the detection precision of our system, and we study the impact of occlusion on the results. Through this evaluation, we assess that our infrastructure achieves 60-cm positioning accuracy in real-world scenarios, providing accurate detection timing and therefore making it usable for collision detection.