Gesture recognition is an important human- computer interaction interface. This paper introduces a novel hand gesture recognition system based on Leap Motion gen. 2. In this system, a spatial fuzzy matching (SFM) algorithm is firstly presented by matching and fusing spatial information to construct a fused gesture dataset. For dynamic hand recognition, an initial frame correction strategy based on SFM is proposed to fast initialize the trajectory of test gesture with respect to the gesture dataset. A notable feature of this system is that it can run on ordinary laptops due to the small size of the fused dataset, which accelerates the calculation of recognition rate. Experiment results show that the system recognizes static hand gestures at recognition rates of 94%-100% and over 90% of dynamic gestures using our collected dataset. This can greatly enhance the usability of Leap Motion.

The Pantograph-catenary system is crucial for transferring electrical power from catenary lines to electrified train and the occurrence of arcing could damage railway operations, thus, it is important to detect arcing. Detecting arcing in complex scenes and detecting different size and shape of arcing is still a challenge. To overcome these issues, a robust image-based semantic segmentation model named ArcMSF is proposed for arcing detection of pantograph-catenary, which designs a novel hybrid multi-scale feature fusion model that aggregates Transformer with CNN to realize arcing pixel segmentation. A down-top decoder for combining low-level features with high-level features is designed to achieve multi-scale level arcing feature detecting in complex scenes. Inspired by the arcing image properties that arcing is always the brightest, the global max features and global threshold features are designed to augment the arcing features. Experiments on dataset IVAIS-PCA2021 and comparative experiments are conducted to demonstrate the effectiveness of the ArcMSF, which can achieve 89.13% segmentation accuracy and the fast inference speed of 23.84ms. Moreover, the detection results have a clear depiction of edge details.

As a result of high-voltage electricity and outdoor environment in high-speed railway traction substation, insulator surface defects are inevitable. However, most of the current unsupervised learning visual detection methods can only achieve patch-level defects detection. To achieve the goal of defects pixel-level segmentation, a universal three-step outdoor support insulators surface defects segmentation framework is proposed to segment insulators surface defects in this article. In this framework, outdoor support insulators are first located using the robust SCRDet network, which can locate the insulators accurately toward various angles. Then, a resolution-improved insulator reconstruction network (RIIR-Net) is proposed to reconstruct resolution-improved nondefect insulators. Finally, defective areas of insulators, such as missing, porcelain drop, and flashover, can be segmented via the proposed coarse-to-fine defects decision strategy, which can postprocess the residual map between the original defective insulator and the reconstructed nondefect insulator to effectively filter noise and segment defects based on decision rules of color feature, area, and HuMoment. Experiments on dataset collected from the Heishan traction substation demonstrate that the newly designed RIIR-Net can indeed improve the quality of reconstructed normal insulators, and the proposed framework can accurately achieve the task of outdoor support insulators defects pixel-level segmentation, including but not limited to missing, porcelain drop, and flashover.

Stable and reliable operation of high-speed railway requires continuous and reliable power supply of traction substation, and isolating switch is one of the most important electrical devices in high-speed railway traction substation. In this paper, to address the issue of isolating switch accurate localization and state recognition simultaneously, we present an automatic isolating switch segmentation and state recognition framework called ISSSR-Net using multitask learning that consists of two stages. First, an isolating switch segmentation network called ISS-Net is proposed for isolating switch pixel-level segmentation precisely, of which a new structure containing a strip pooling module, a channel attention and three pyramid pooling modules is designed to greatly improve the segmentation and recognition performance, even in complex conditions such as rain, snow and fog. Second, to improve state recognition accuracy, the segmentation map yielded from the ISS-Net and the feature map from the shared backbone are together fed into isolating switch recognition network called ISR-Net to recognize its three states. In addition, a global context block is integrated into ISR-Net to further improve state recognition accuracy. Extensive experimental results on a self-collected dataset from Heishan traction substation corroborate that this paper provides an effective and robust method to achieve isolating switch segmentation and state recognition simultaneously. The M IoU and ${F}~1$ -score of segmentation reach 0.93 and 0.94 respectively, which is better than U-Net and its variants, and other SOTA. The ${F}~1$ -score of recognition reach 1.00, which is higher than HOG+SVM and other deep learning methods that have been experimented.

Monitoring train operation status is one of the most important tasks for ensuring rail operation safety. Pantograph and catenary (PAC) are collecting systems of the electric current from traction power supply system, and the stability of the contact between pantograph and catenary guarantees the stable power. However, most existing contact point (CPT) detection methods are always difficult to achieve precise positioning results, especially in complicated background. This paper proposes a novel fast and accurate contact point detection method based on multiple strategies, which combine three modules. First, an improved kernelized correlation filter model in real-time tracking module was adopted to track the contact region. Then the pixel-level detection module was used to detect contact point in contact region via the proposed contact point regression residual network (CPRR-Net). Finally, a filter-based optimization module was added to correct the contact position using the Kalman filter. This work additionally employed a new rail dataset PAC-TPL2020 to prove the effectiveness and feasibility of the proposed multiple strategies in real-world scenarios, and the experimental results demonstrated the robustness and high accuracy (97.07% within 3 pixels and 99.97% within 5 pixels) of our model. It is noteworthy that our mothed runs at 65 frames per second for monitoring PAC contact points.