POT-YOLO: Real-Time Road Potholes Detection Using Edge Segmentation-Based Yolo V8 Network

Detecting and avoiding potholes is a very challenging task in India due to the poor quality of construction materials used in road privilege systems. Identifying and repairing potholes as soon as possible is crucial to preventing accidents. Roadside potholes can cause serious traffic safety problems and damage automobiles. In this article, a novel pothole detection using Yolov8 (POT-YOLO) has been introduced for detecting the types of potholes such as cracks, oil stains, patches, and pebbles using POT-YOLOv8. Initially, pothole videos are converted into frames of images for further processing. To reduce distortions, these frames are preprocessed with the Contrast Stretching Adaptive Gaussian Star Filter (CAGF). Finally, the preprocessed images are identifying the region of pothole using Sobal edge detector and detect the pothole using YOLOv8. The POT-YOLO approach was simulated with Python code. The simulation result demonstrate that the POT-YOLO methods performance was measured in terms of ACU, PRE, RCL, and F1S. The POT-YOLO achieves an overall ACU of 99.10%. Additionally, POT-YOLO model achieves 97.6% precision, 93.52% recall, and 90.2% F1-score. In comparison, the POT-YOLOv8 network improves the ACU range than the existing networks such as Faster RCNN, SSD, and mask R CNN. The POT-YOLO approach improves the overall ACU by 12.3%, 0.97%, and 1.4% better than ML-based DeepBus, automatic color image analysis using DNN and ODRNN, respectively.