| DOI | Resolve DOI: https://doi.org/10.1109/VTC2021-Spring51267.2021.9448887 |
|---|
| Author | Search for: Mammeri, Abdelhamid1; Search for: Siddiqui, Abdul Jabbar1; Search for: Zhao, Yiheng1 |
|---|
| Affiliation | - National Research Council of Canada. Automotive and Surface Transportation
|
|---|
| Format | Text, Article |
|---|
| Conference | 2021 IEEE 93rd Vehicular Technology Conference (VTC2021-Spring), April 25-28, 2021, Helsinki, Finland [Virtual Conference] |
|---|
| Subject | railway track segmentation; UAV imagery; UAV-based inspections |
|---|
| Abstract | In the railway sector, track inspections are regularly needed to monitor the track conditions for potential hazards in order to ensure safety and security of life and property. Recently, conducting infrastructure inspections and monitoring using UAVs has gained attention in various industries including the railways. The rapid development of advanced deep learning and machine vision techniques have given rise to automated railway hazard detection systems based on UAV-based imagery. A major task in such systems is to localize or segment the railway tracks in UAV-based images. This paper investigates the effectiveness of a fully convolutional encoder-decoder type segmentation network called U-Net for the task of segmenting rail track regions from UAV-based images. Through experimental evaluations using a proprietary real-world dataset, we demonstrate U-Net's effectiveness in terms of mean Intersection over Union (IoU). Such methods of rail track segmentation are particularly useful in applications such as automated UAV navigation along rail tracks. |
|---|
| Publication date | 2021-06-15 |
|---|
| Publisher | IEEE |
|---|
| In | |
|---|
| Language | English |
|---|
| Peer reviewed | Yes |
|---|
| Export citation | Export as RIS |
|---|
| Report a correction | Report a correction (opens in a new tab) |
|---|
| Record identifier | 60f96967-dbb3-42ca-9c74-6a0818071318 |
|---|
| Record created | 2021-12-09 |
|---|
| Record modified | 2021-12-09 |
|---|
