The regular detection of pavement cracks is critical for life and property security. However, existing deep learning-based methods of crack detection face difficulties in terms of data acquisition and defect counting. An automatic intelligent detection and tracking system for pavement cracks is proposed. Our system is formed of a pavement crack generative adversarial network (PCGAN) and a crack detection and tracking network called YOLO-MF. First, PCGAN is used to generate realistic crack images, to address the problem of the small number of available images. Next, YOLO-MF is developed based on an improved YOLO v3 modified by an acceleration algorithm and median flow (MF) algorithm to count the number of cracks. In a counting loop, our improved YOLO v3 detects cracks and the MF algorithm tracks the cracks detected in a video. This improved algorithm achieves the best accuracy of 98.47% and F1 score of 0.958 among other algorithms, and the precision-recall curve was close to the top right. A tiny model was developed and an acceleration algorithm was applied, which improved the detection speed by factors of five and six, respectively. In on-site measurement, three cracks were detected and tracked, and the total count was correct. Finally, the system was embedded in an intelligent device consisting of a calculating module, an automated unmanned aerial vehicle, and other components.

Conventional algorithms are not sensitive to small objects like pavement cracks. We developed a pavement crack detection method based on a convolutional neural network (CNN) with multiple feature layers. The model extracts multi-scale features to increase the accuracy of pavement crack recognition. After hyperparameters tuning, the model accuracy reached 98.217%, and the detection rate reached 96.6 frame per second (FPS). These results showed that the model could be feasibly used for real-time crack detection. Using multiple aspect ratio anchor boxes and multi-scale feature maps, the accuracy can be improved by 1.809% and 5.016%, respectively. Compared with the traditional detection algorithm, our model was optimal in terms of F1 score and Precision-recall curve, and it was less affected by shadows and road markings and detected the crack boundaries more accurately. An on-site crack detection experiment was carried out to quantify the effectiveness of the model in crack detection.