This article presents an autonomous navigation approach based on a transmission tower for unmanned aerial vehicle (UAV) power line inspection. For this complex vision task, a perspective navigation model, which plays an important role in the description and analysis of the flight strategy, is introduced. Based on the proposed navigation model, valuable cues are excavated from a perspective image, which enhances the capability of the perception of three-dimensional direction and simultaneously improves the safety of intelligent inspection. Specifically, for robust and continuous localization of the transmission tower, a developed detecting-tracking visual strategy-comprised tower detection based on a faster regionbased convolutional neural network and tower tracking by kernelized correlation filters-is presented. Further, segmentation by fully convolutional networks is applied to the extraction of transmission lines, from which the vanishing point (VP), an important basis for determining the flight heading, can be obtained. For more robust navigation, the designed scheme addresses the scenario of a nonexistent VP. Finally, the proposed navigation approach and constructed UAV platform were evaluated in a practical environment and achieved satisfactory results. To the best of our knowledge, this article marks the first time that a navigation approach based on a transmission tower is proposed and implemented.

Semi-supervised anomaly detection identifies abnormal (testing) observations which are different from normal (training) observations. In many practical situations, anomalies are poorly insufficient and not well defined, while the normal data are easily sampled, have a wide variety, and may not be classified. For this paradigm, we propose a novel end-to-end deep network as an anomaly detector only trained on normal samples. Our architecture consists of a conditional variational auto-encoder (CVAE), a feature discriminator (FD), and an adversarially trained WGAN-GP discriminator. The CVAE is designed as a generator to reconstruct images. It leverages underlying category information and multivariate Gaussian distributions to regularize the latent space, enabling a smooth and informative manifold. For anomalies which have a certain similarity to normal data, we perform active negative training by generating potential outliers from the latent space to limit network generative capability. In order to capture data characteristics, we maximize the mutual information between the inputs and the latent codes by the FD. It enhances the relationship between the high-dimensional image space and corresponding encoded vectors. To promote reconstruction, a structural similarity loss is applied to robustly recover local texture details and the WGAN-GP discriminator is employed to aid in generating photo-realistic images. We distinguish anomalies by computing a reconstruction-based anomaly score. Different from recent encoder-decoder or GAN-based architectures, our approach considers input categories, constructs, and exploits a useful manifold in an unsupervised manner and has a stronger reconstruction capability. The experimental results demonstrate that the proposed approach outperforms state-of-the-art methods over several benchmark datasets. INDEX TERMS Semi-supervised anomaly detection, conditional variational auto-encoder, generative adversarial networks, informative manifold, structural similarity loss.

Employing unmanned aerial vehicles to conduct close proximity inspection of transmission tower is becoming increasingly common. This article aims to solve the two key problems of close proximity navigation-localizing tower and simultaneously estimating the unmanned aerial vehicle positions. To this end, we propose a novel monocular vision-based environmental perception approach and implement it in a hierarchical embedded unmanned aerial vehicle system. The proposed framework comprises tower localization and an improved point-line-based simultaneous localization and mapping framework consisting of feature matching, frame tracking, local mapping, loop closure, and nonlinear optimization. To enhance frame association, the prominent line feature of tower is heuristically extracted and matched followed by the intersections of lines are processed as the point feature. Then, the bundle adjustment optimization leverages the intersections of lines and the point-to-line distance to improve the accuracy of unmanned aerial vehicle localization. For tower localization, a transmission tower data set is created and a concise deep learning-based neural network is designed to perform real-time and accurate tower detection. Then, it is in combination with a keyframe-based semi-dense mapping to locate the tower with a clear line-shaped structure in 3-D space. Additionally, two reasonable paths are planned for the refined inspection. In experiments, the whole unmanned aerial vehicle system developed on Robot Operating System framework is evaluated along the paths both in a synthetic scene and in a real-world inspection environment. The final results show that the accuracy of unmanned aerial vehicle localization is improved, and the tower reconstruction is fast and clear. Based on our approach, the safe and autonomous unmanned aerial vehicle close proximity inspection of transmission tower can be realized.