Pavement distresses monitoring on a stretch of NH-44 (India) using dcnn

Automatic and timely detection of pavement distress is important, as it helps in quick repair of pavement distress, that reduces the cost of repairing of damaged road structures, and decreases traffic accidents. Sensor and vision-based modelling cannot match real-time pavement distress detection criteria. To address this shortcoming, a new method, YOLOv8, an upgraded CNN model, is used for high-accuracy real-time identification of pavement distress: potholes, cracks, patches, unevenness, and shoving. Study prepares data using ROBOFLOW and implements YOLOv8 in Google COLAB for real-time detection of pavement distress. The section of NH-44 from Palwal to Agra, along with the surrounding district roads, was selected for modelling in order to analyse and describe the different types of pavement distress. A total of 14,000 images were captured using a high-resolution 64-megapixel camera. Dataset was prepared using ROBOFLOW and seven different classes of pavement distresses are formed: Alligator-Block Crack, Delamination, Longitudinal Crack, Pot Hole, Ravelling, Rutting and Shoving. The aforementioned model gave results for these seven classes, with an average precision (mAP) of 81% at the IOU 50 level and mAP50-95 64.4% for alligator cracking. Similarly, a pothole has 65.6% mAP50 and 34.8% mAP50-95, but the model has a very low accuracy, in terms of detection, in the event of rutting, with 16.6% at mAP50 and 6.6% at mAP50-95. Present study will be of great help to ministry of road transport and highways (MORTH) by providing automated detection, improved accuracy, real time monitoring, and cost saving in maintenance of pavement distresses.