TTPLA: An Aerial-Image Dataset for Detection and Segmentation of Transmission Towers and Power Lines (2010.10032v1)

Abstract: Accurate detection and segmentation of transmission towers~(TTs) and power lines~(PLs) from aerial images plays a key role in protecting power-grid security and low-altitude UAV safety. Meanwhile, aerial images of TTs and PLs pose a number of new challenges to the computer vision researchers who work on object detection and segmentation -- PLs are long and thin, and may show similar color as the background; TTs can be of various shapes and most likely made up of line structures of various sparsity; The background scene, lighting, and object sizes can vary significantly from one image to another. In this paper we collect and release a new TT/PL Aerial-image (TTPLA) dataset, consisting of 1,100 images with the resolution of 3,840$\times$2,160 pixels, as well as manually labeled 8,987 instances of TTs and PLs. We develop novel policies for collecting, annotating, and labeling the images in TTPLA. Different from other relevant datasets, TTPLA supports evaluation of instance segmentation, besides detection and semantic segmentation. To build a baseline for detection and segmentation tasks on TTPLA, we report the performance of several state-of-the-art deep learning models on our dataset. TTPLA dataset is publicly available at https://github.com/r3ab/ttpla_dataset

• The paper presents TTPLA, a dataset with 1,100 high-resolution images and 8,987 annotated instances to enhance instance segmentation research.
• The authors employ diverse eye-level, front, and side view strategies under varied lighting to capture real-world aerial imagery challenges.
• The study evaluates Yolact with Resnet backbones, revealing low average precision scores of 22.96% for bounding boxes and 15.72% for masks, pointing to segmentation challenges.

An Overview of TTPLA: An Aerial-Image Dataset for Detection and Segmentation of Transmission Towers and Power Lines

The paper introduces the TTPLA dataset, a novel contribution to the field of computer vision, specifically focusing on accurately detecting and segmenting transmission towers (TTs) and power lines (PLs) from aerial images. The motivation for this work is founded on the pressing need for automated inspection and monitoring of power grids to enhance security and efficiency. Traditional methods involving human inspectors or robotic systems present challenges related to cost and accuracy, thereby making UAV-based solutions desirable.

Key Contributions and Methodology

1. Dataset Creation and Characteristics:
   • The TTPLA dataset comprises 1,100 high-resolution images (3,840×2,160 pixels) with manual annotations for 8,987 instances of TTs and PLs.
   • This dataset captures the challenges inherent in aerial images, such as varying lighting conditions, scene complexity, and object scale imbalance.
   • TTPLA supports instance segmentation, a more granular annotation providing separate masks for each object instance, unlike traditional semantic segmentation datasets.
2. Data Collection Strategies:
   • A variety of angle views, including top, front, and side perspectives, are incorporated to improve model robustness.
   • The collection was executed in different states at various times to encompass a range of backgrounds and lighting conditions.
   • The manual annotation was performed using polygons to ensure precise boundary delineations in challenging conditions where PLs might blend with backgrounds.
3. Baseline Model and Evaluation:
   • State-of-the-art segmentation models, specifically Yolact with Resnet backbones, were evaluated on TTPLA.
   • Average precision scores were used for performance measurement, with results showing significant room for improvement (AP of 22.96% for bounding boxes and 15.72% for masks).
   • Issues such as classification falseness and detection errors were identified, indicating the complexity of aerial imagery and the current limitation of segmentation models in effectively handling such scenarios.

Implications and Future Work

The introduction of TTPLA marks a substantial step forward in UAV-based power grid inspection, offering a valuable resource for the development of advanced computer vision models. This dataset is expected to facilitate the training of more robust models capable of real-time operation in complex environments with high precision demands.

Practical implications extend to improved UAV navigation and obstacle avoidance, contributing to safer and more efficient autonomous operations in utilities and beyond. The real-world complexities captured in TTPLA also offer theoretical insights into instance segmentation challenges, encouraging the exploration of new architectures or learning paradigms that can address the issues of overlapping instances and small object detection.

Future research may involve leveraging synthetic data augmentation, novel loss function designs, or domain adaptation techniques to mitigate the current dataset challenges. Additionally, advancements in learning algorithms, possibly through unsupervised or semi-supervised methods, could further refine the models trained on TTPLA.

In summary, the TTPLA dataset stands as a significant academic resource for accelerating research in aerial imagery analysis, particularly in the critical application domain of power infrastructure inspection. Its development paves the way for deeper exploration and innovation in segmentation tasks, promising notable enhancements in both safety and efficiency of inspection processes.