Real-Time Vision-Based System of Fault Detection for Freight Trains

A Unified Framework for Fault Detection of Freight Train Images Under Complex Environment

“…[c * > 0] is an indicator that equals to 1 if the argument is true and 0 otherwise. Besides that, all local features are pre-computed before multi-RPN and detection without redundant computation [4]. The effectiveness of multi-RPN will be further described in the following Section IV-B.…”

“…To this end, we firstly evaluate our proposed light-weight backbone on three datasets, ImageNet ILSVRC 2012 [31], PASCAL visual object classes (VOC) 2007 [32] and MS COCO [33]. Then we compare the proposed framework with state-of-the-art fault detectors and well-known object detection methods on six fault datasets [4], [7]. We conduct all of our experiments using Caffe [34] on a single NVIDIA GeForce GTX1080Ti GPU.…”

“…2) Fault Datasets: To evaluate the performance of our method, six fault datasets [4], [7] for freight train images are directly used in this study, including angle cock, bogie block key, brake shoe key, cut-out cock, dust collector, and fastening bolt on brake beam. Some typical samples of freight train images are shown in Fig 1(b).…”

“…But it is insufficient to achieve fast speed, and its model size is huge (>550 MB). Based on the above FTI-FDet, Zhang et al [4] proposed a specialized light fault detector (Light FTI-FDet) which pursues a balance between accuracy and speed. However, its model size is still over 89 MB.…”

Online Pantograph-Catenary Contact Point Detection in Complicated Background Based on Multiple Strategies

“…Consequently, it is of great significance to propose a novel method to accurately identify and locate the contact point for further research on PAC system fault diagnosis. In recent years, deep learning-based methods have been widely applied for railway defect detection [10] [11], [12] and greatly improved the accuracy and stability of surveillance methods.…”