Defects Detection & Condition Assessment of Road Infrastructure Using Deep Learning Algorithm

Road surface conditions worsen owing to vehicle overburden, freak accidents, low quality road development, and other atmospheric conditions. It may lead to road debris which may give rise to vehicle damages, traffic jams and creates obstacles to civil administration in identifying such defects accurately and mending it. Manual inspection of road defects (i.e., bumps, potholes, cracks, speed breakers etc.,) is time consuming, expensive and ineffective solution hence, automation is achieved using advanced algorithms. In this paper, real-time road condition assessment and defects’ detection are performed using captured images. The prototype comprised of camera module V2 mounted on the front of vehicle along with GPS module to collect corresponding location information and integrated with Raspberry Pi 3 B+. The training model is designed based on Convolution Neural Network (CNN) for object detection algorithm. Transfer learning using the object detection dataset of Common Object in Context (COCO) is used to enhance the performance and saves time. The dataset of 400 images having resolution 720 x 720 are included among which 200 images are healthy and 200 are unhealthy. The performance is improved by tuning the model and collected road images are trained to detect the road faults with 98 % accuracy. The identified defects along with their location parameters are stored in cloud for road authorities to be used for road planning and maintenance. The proposed system is validated with manual and accelerometer sensor detection where the sensor is tested by deploying it at axle using Axle Based Acceleration (ABA) method. The validation shows that the proposed system efficiently detects road faults automatically with maximum accuracy.