Potholes Detection Using Optimal Object Detection Model in Real-time for Different Use Cases

Well-maintained roads play a vital role in any country’s economic growth as the primary mode of transportation. One of the critical challenges in developing or underdeveloped countries is to regularly evaluate the road condition for timely repair as it’s time-consuming & costly with the traditional manual process to identify the potholes. Additionally, for autonomous vehicles, it would be difficult to navigate without significantly high accuracy & precision along with low latency for appropriate object detection.This research uses various CNN-based YOLO models (YOLOv5-8 with pre-trained weights) to identify the best possible model based on the use cases. According to the analysis of outcomes and performance in this research, the YOLOv7x model achieved the highest efficiency. The performance of this model has exceeded most of the relevant and recent research results. Apart from that, some other YOLO models are recommended based on applications or device types.With this higher precision(mean average precision-mAP) & faster inference time for detecting potholes, it will be helpful for speedier road surveys and boost the autonomous vehicle industry. Apart from the scenarios mentioned above, warning signals broadcast to the nearest vehicles in runtime will also be possible to avoid accidents due to the worst potholes.