4.3 Article

The NITRDrone Dataset to Address the Challenges for Road Extraction from Aerial Images

Publisher

SPRINGER
DOI: 10.1007/s11265-022-01777-0

Keywords

AI; aerial image; Semantic segmentation; CNN; UAV

Ask authors/readers for more resources

Recent years have seen significant advancements in small-scale remote sensors such as UAVs, particularly in the field of computer-vision tasks like aerial image segmentation. This paper introduces the NITRDrone dataset, which focuses on extracting road networks from aerial images captured at different locations on the NITR campus. Extensive experiments have been conducted to validate the dataset's effectiveness, with U-Net achieving the best performance. The availability of the NITRDrone dataset is expected to boost research and development in visual analysis of UAV platforms.
Recent years have witnessed a dramatic evolution in small-scale remote sensors such as Unmanned aerial vehicles (UAVs). Characteristics such as automatic flight control, flight time, and image acquisition have fueled various computer-vision tasks, providing better efficiency and usefulness than fixed viewing surveillance cameras. However, in constrained scenarios, the number of UAV-based aerial datasets is still low, which comparatively focuses on specific tasks such as image segmentation. In this paper, we present a high-resolution UAV-based image-dataset, named NITRDrone focusing on aerial image segmentation tasks especially extracting the road networks from the aerial images. The images and video sequences in this dataset are captured over different locations of the NITR campus area, covering around 650 acres. Thus, it provides many diversified scenarios to be considered while analyzing aerial images. In particular, the dataset is prepared to address the existing challenges in UAV-based aerial image segmentation problems. Extensive experiments have been conducted to prove the effectiveness of the proposed dataset to address the aerial segmentation problems through the existing state-of-the-art methodologies. Out of the considered baseline methodologies, U-Net performs the best with an intersection of union (IoU) of 0.77, followed DeepLabplusException (IoU: 0.74) and SegNet (IoU: 0.68). We hope the NITRDrone dataset will encourage the researchers while boosting the research and development in the visual analysis of UAV platforms. The NITRDrone dataset is available online at: [haps://github.coirildrone-vision/NITRDrone-Dataset].

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.3
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available