|
Proposing a Deep Learning-Based Method for Estimating the Absolute Position of Potholes in Urban Streets
|
Homayoon Hadigol *   , Ali Hosseininaveh |
|
|
|
Abstract: (7 Views) |
| Accurate identification and location of potholes on asphalt surfaces plays an important role in improving driver safety, reducing maintenance costs, and optimizing the management of urban transportation infrastructure. In this study, a deep learning-based method is presented that achieves these goals using affordable and readily available technologies. The proposed method uses a deep neural network to detect and segment potholes in images captured by smartphone cameras. Then, using a data fusion algorithm, the information extracted from the images is combined with Global Positioning System (GPS) and Inertial Measurement Unit (IMU) data received from the phone sensors to estimate the absolute location of the potholes in the Universal Coordinate System (UTM) with high accuracy. This innovative approach effectively overcomes the challenges associated with converting image coordinates to universal coordinates and errors caused by the low accuracy of GPS data. In order to evaluate the performance of the proposed method, a comprehensive dataset of urban streets was collected using a common smartphone. In this process, the location of potholes was determined in two ways: once using the proposed method presented in this study and again using a G1 Plus Sout multi-frequency receiver with an accuracy of 0.012 m, which was considered as an accurate reference. Then, the results of these two methods were compared with each other to evaluate the accuracy and efficiency of the proposed method. The experimental results show that this method is able to estimate the absolute location of potholes with an average error of less than 2 m in the UTM coordinate system. This approach allows for accurate and up-to-date mapping of the surface condition of urban streets using everyday and publicly available tools. In addition, the simplicity and cost-effectiveness of this method allows for widespread citizen participation in data collection and improvement of urban infrastructure. |
Article number: 4 |
| Keywords: Deep Learning, Neural Network, Object Detection, Visual Odometry, Data Fusion |
|
|
Full-Text [PDF 875 kb]
(7 Downloads)
|
Type of Study: Research |
Subject:
Photo&RS
Received: 2025/02/24
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
