Pedestrian symbols inherently have specific relative positions concerning one another, as perceived by individuals without visual impairments. Developing a vision-impaired assistive device capable of detecting these relative positions of pedestrian objects could effectively guide users in the correct direction when crossing pedestrian roads. For instance, it is commonly observed that pedestrian lights are positioned above pedestrian lanes. This study demonstrated this concept by employing multiple object recognition architectures, such as YOLOv8 and Faster RCNN, to detect the presence of pedestrian lights in green or “go” status and the existence of pedestrian lanes. YOLOv8 performed better with 88.8% precision and 90.8% recall than Faster-RCNN with only 83.6% precision and 86.2%, respectively. Furthermore, the relative positions of these elements are estimated using the centers of the generated proposals or bounding boxes. Several facing direction scenarios are illustrated using the developed difference method that systematizes the recognition of object cues for a person with normal vision. By integrating this into a visually-impaired assistive device, it is now possible to adjust the user's facial direction, thus ensuring safe pedestrian crossing.