Zhenzhou Fu

Electric locomotives depend on their roof power supply systems, which are threatened by unknown foreign objects due to the complex outdoor environment. Timely and accurate detection of these objects is vital for train safety. Existing methods are limited by the diversity and unpredictability of foreign objects, difficulty in collecting abnormal samples, and poor image consistency. To address these challenges, this paper proposes a Patch-Wise Neighborhood Feature Distribution Modeling (PNFDM) algorithm for anomaly detection. It uses multiple multivariate Gaussian distributions to model normal image features at the patch level and computes Mahalanobis distances to generate an anomaly heatmap. A patch-wise neighborhood feature sharing strategy improves the signal-to-noise ratio of the anomaly heatmap. To reduce storage requirements, a low-rank approximation using SVD decomposition reduces model size by nearly 110 times and speeds up loading by 103 times. A chi-square distribution-based normalization remaps the Mahalanobis distance to a 0-1 range, facilitating threshold selection in anomaly localization. Experiments show that this method significantly outperforms others on a locomotive roof anomaly dataset and generalizes well to the public MVTec AD dataset. The approach allows for rapid model updates without needing abnormal samples, training on 100 samples in just 39.76 seconds. Dataset and code are available at: https://github.com/zzfu-buaa/PNFDM.

Train fault detection often relies on comparing collected images with reference images, making accurate image type recognition crucial. Current systems use Automatic Equipment Identification (AEI) devices to recognize carriage numbers while capturing images, but damaged Radio Frequency (RF) tags or blurred characters can hinder this process. Carriage linear array images, with their high resolution, extreme aspect ratios, and local nonlinear distortions, present challenges for recognition algorithms. This paper proposes a method tailored for recognizing such images. We apply an object detection algorithm to locate key components, simplifying image recognition into a sparse point set alignment task. To handle local distortions, we introduce a weighted radial basis function (RBF) and maximize the similarity between Gaussian mixtures of point sets to determine RBF weights. Experiments show 100% recognition accuracy under nonlinear distortions up to 15%. The algorithm also performs robustly with detection errors and identifies categories from 79 image classes in 24 ms on an i7 CPU without GPU support. This method significantly reduces system costs and advances automatic exterior fault detection for trains.

Train fault detection primarily relies on the comparative analysis of collected images with standard reference images, making the accurate identification of the collected image types crucial for effective fault detection. Most current systems achieve indirect classification of carriage images by using Automatic Equipment Identification (AEI) devices to recognize carriage identification number information while capturing the carriage images. However, damaged RF tags or blurred characters can hinder these devices, obstructing further train fault analyses. The unique features of carriage linear array images, such as high resolution, extreme aspect ratios, and local non-linear distortions, pose significant challenges to recognition algorithms. This paper proposes a method specifically designed to recognize these types of images. First, we implement an object detection algorithm to locate the positions of key components in carriage linear array images. By leveraging the spatial layout mapping between these key components and the carriage type classification, we simplify the complex image recognition problem into a sparse point set alignment task. To ensure effective registration despite local distortions in the linear array images, we propose a weighted radial basis function. To address the issues of unknown matching relationships and quantity mismatches during point set alignment, the proposed objective function is designed to maximize the similarity between Gaussian mixtures of point sets, thereby determining the RBF weights. Extensive experiments have demonstrated that our algorithm achieves 100% recognition accuracy under local nonlinear scale distortions of less than 15%. The algorithm remains effective even in the presence of some false positives or missed detections in key component object detection results. It can accurately identify the target category from 79 image categories within 24 milliseconds on an i7 CPU without GPU support. It is of great significance to reduce system cost and promote automatic exterior fault detection for train rolling stock.

Abstract.In pushbroom-style train imaging systems, the efficiency and accuracy of image alignment are crucial for improving train fault detection accuracy. However, nonlinear scale distortion in linear array images poses significant challenges to alignment precision. To address this, our study introduces an innovative image alignment algorithm for linear arrays, adept at handling nonlinear scale distortions. This algorithm is particularly effective in aligning heterogeneous images, even with substantial differences in texture features. The developed dynamic step-length sliding window strategy, feature point matching using geometric constraints, polynomial-constrained outlier elimination, and interval feature matching fusion significantly enhance both the accuracy and density of feature point matching. Further, the application of weighted radial basis function (WRBF) facilitates precise coordinate transformation in the image remapping process. Comprehensive experimental evaluations demonstrate the algorithm’s superior alignment precision and efficiency in both homogenous and heterogeneous image alignment scenarios, markedly boosting train fault detection accuracy. The algorithm’s versatility extends its utility beyond train fault detection to broader applications in pushbroom-style imaging system alignment tasks.

The automatic inspection of real-time pantograph-and-catenary (PAC) operational status is important for ensuring the safety of the railway power supply system and realizing its intelligent operation and maintenance. However, state-of-the-art contact-loss monitoring methods cannot achieve high precision, high speed, and robust effects under complex conditions, such as frequent line changes, multi-suspended line interference, unknown complex backgrounds, and brightness variations. Therefore, a new two-stage online detection method for PAC operational status is proposed. First, we propose a three-module-assisted high-precision key point positioning detector for all possible intersection points of PAC. It can realize end-to-end PAC contact corner points detection for double branch contact mode in the process of line changes, instead of the original way of tracking and then detecting. Each auxiliary module adopts a minimalist ${3\times 3} $ convolutional layer without pre-training to enhance the feature expression ability of the corresponding layer in the backbone, thus improving the sub-pixel detection accuracy. Second, we propose a topological accuracy compensation algorithm to compensate for the mismatch error between the corner points in the image and the projection of the actual spatial contact point owing to the self-occlusion of PAC, thus improving the positioning precision of the real 3-D contact point. Physical simulation, accuracy verification, and field experiments prove that our PAC status inspection algorithm can achieve accurate conductor height and stagger value measurements under complicated environmental interference, thus ensuring the safety of train operation.

As an important part of the online monitoring in electrified railway, pantograph operational pose inspection plays a vital role in pantograph and catenary system (PCS) security. Due to the impact of the complex conditions on the train roof, it is difficult to replace the existing video surveillance with fully intelligent operational pose inspection. Therefore, this article proposes a novel accurate monocular 3-D pantograph horn edge-based pantograph pose inspection method in the power of superior ability of deep learning. It mainly includes two steps: edge detection and pose optimization. First, we highlight a brand-new high-precision subpixel edge prediction network. The downsampling heatmap and offset prediction map in our edge detector are joined to predict full-size edge points and thus improve the accuracy to the subpixel level. In addition, our newly proposed numerical gradient smoothing loss function overcomes the “jaggy” phenomenon of the edge curves. Second, we propose a nonpoint-to-point corresponding 3-D pose measurement method of the monocular camera. In the iterative optimized process, we innovatively apply the chamfer distance as an optimization function to dynamically adjust the correspondence between the reprojection points and the predicted edge points to optimal convergence. Ablation and comparative experiments demonstrate that our method achieves excellent performance (F-score 93.0% in FEP _2.0 (fixed edge probability) with 82 frames per second (FPS), overcoming the complex conditions of the practical online monitoring sequences. The actual pose measurement results further confirm our efficiency, thereby achieving a robust real-time and high-precision monocular 3-D pantograph operating pose inspection.

Good contact conditions between the pantograph and catenary ensure the safety of electrified railways. However, the relevant state-of-the-art status inspection methods cannot achieve high speed and robust effects under complex conditions, such as multisuspended line interference, unknown complex backgrounds, and light and dark changes. Therefore, a two-module contact status diagnosis methodology, consisted of long-term tracking and 3-D reconstruction, is proposed in this study. This long-term tracking method is a combination of tracking, detection, and discrimination, focusing on locating the contact area of high-speed movement in complex environments. First, a brand-new tracking paradigm, namely, GMSTrack, based on feature matching is developed to adapt to complex background changes. Second, we modify the loss function via GIoU-plus loss for box regression and adaptive wing loss for landmark regression on CenterNet and realize a dramatic improvement in detection accuracy at 125 FPS. Third, sequential information is collected for the transformer-based discriminator to evaluate the correctness of the current tracking result. Experimental results show that the optimal strategy that combines tracking, detection, and discrimination can locate the contact area in complex environments with an accuracy higher than 93% and with a speed higher than 84 FPS. Finally, dynamic stagger value and height of conductor that can reflect contact status are analyzed through the 3-D reconstruction of the binocular contact points. The overall status inspection methodology can be applied to the on-site high-speed trains at speeds exceeding 80 FPS.