Development of UAV-based pavement crack identification system using artificial intelligence

Ersöz, Ahmet Bahaddin
Building an accurate, robust and timely working Pavement Crack Identification System (PCIS) is one of the challenging components of Pavement Management Systems (PMSs). The ultimate aim of PCIS is to have autonomous inspection methods integrated into PMS. This way a modern PCIS may replace the currently used methods to eliminate their shortcomings such as being labor intensive, biased and time consuming. With the recent introduction of Unmanned Aerial Vehicles (UAVs), engineering research studies are inclined towards their use in various applications. In this study, UAVs are employed to capture the images of the pavement surface, from which pavement cracks are detected using digital image processing techniques and classified with a machine learning algorithm called Support Vector Machines (SVMs). The proposed pavement crack identification method using images includes preliminary operations, making the images uniformly illuminated and noise free. Comparatively darker regions in pre-processed images called connected components are obtained using automated thresholding. Through geometric features extracted from connected components, SVMs are used to classify the cracks, through which the connected components are classified into four groups: longitudinal cracks, transverse cracks, alligator cracks and non-crack regions. A case study was performed to measure the performance of the proposed method. The crack prediction results were quite successful. The proposed PCIS has a few benefits such as being cheap and computationally efficient and therefore, it can be used in practical pavement management applications successfully.