Advanced pavement distress recognition and 3D reconstruction by using GA-DenseNet and binocular stereo vision

Pavement distress detection is the first and most important step in pavement maintenance procedures. The detection results directly determine the maintenance mileage and strategy. Traditional detection methods are either inaccurate or expensive. It is necessary to develop an accurate and inexpensive pavement distress detection method. This study presents an advanced pavement distress detection method based on deep learning and binocular stereo vision. The Genetic Algorithm (GA)-DenseNet is used to classify more than 6500 distress datasets to identify both two-dimensional (2D) and three-dimensional (3D) distresses. The input image size of the model is 224 * 224 pixels, and the recognition speed is 0.33 s per piece. Binocular stereo vision and point cloud processing are used for 3D reconstruction and extracting morphology features of potholes. The test results show that the average accuracy of depth and area is 98.9% and 98.0%, respectively. The RGB image is first used to calculate the pavement damage rate (DR) and pavement surface condition index (PCI) of road pothole defect in millimeter-level accuracy, and the overall detection accuracy is 88.2%. This study demonstrates a promising inexpensive way to automatically detect 3D pavement distresses, thereby providing direct guidance in pavement maintenance decision-making.

Automated summary

This study presents an advanced pavement distress detection method based on deep learning and binocular stereo vision. The Genetic Algorithm (GA)-DenseNet is used to accurately classify two-dimensional and three-dimensional distresses. Using binocular stereo vision and point cloud processing, 3D reconstruction and morphological feature extraction of potholes are performed. Test results show high accuracy in depth and area detection. The study demonstrates a promising and inexpensive way to automatically detect 3D pavement distresses, providing guidance for pavement maintenance decision-making.