Are Utility Poles Within Sight? Unleashing the Power of Computer Vision for Effortless Roadside Management

We present an approach to support digitalization of roadside objects, specifically utility poles, which are vital infrastructure components for power distribution, telecommunications, and other services. The proposed approach aims to streamline and improve the management of utility poles by offering a computer vision-based method for their automatic localization and mapping. To this aim, we have developed a novel method for deep learning-based vision for automatically localizing and mapping roadside utility poles. We trained a roadside infrastructure detection model based on the YOLOv8 object detector using the large-scale road-view Mapillary Vistas Dataset. The localization part is performed with the self-supervised depth-estimation network MonoViT, which estimates the distance between the camera and the observed utility pole from the input single image. Next, the generalization ability of the proposed approach is tested using our ZED custom dataset. The test results show that our approach achieves a high detection recall score of 82.7% on our custom dataset and a mean error rate of 6.31% in locating utility poles. The proposed method supports new applications in the fields of intelligent transportation system and autonomous driving research.