Asphalt pavement crack localisation based on depth camera and image-processing algorithms

Under heavy traffic loads, pavement cracks have become a major factor affecting asphalt pavement performance, while conventional rehabilitation remains time-consuming and labour-intensive. Intelligent maintenance equipment offers an effective way to improve repair efficiency. Although deep learning methods have been widely used for crack detection, they often require large annotated data sets, high computational resources and retraining for different pavement conditions. In contrast, traditional image-processing methods are computationally efficient and data-independent, but their accuracy is easily affected by noise and shadows. To address these limitations, this study proposes a depth-camera-based crack localisation and tracking method that integrates multiple image-processing algorithms, including shadow removal, crack mask extraction and skeleton point identification. The proposed method was implemented on a self-developed intelligent asphalt-pavement repair robot equipped with a depth camera. Experimental results show that the method can track pavement cracks in real time during robot movement, achieving an intersection over union value of 0.80, compared with 0.42 for the conventional method. The proposed approach provides a practical balance between detection accuracy and computational efficiency.