Rapid Detection of Tools of Railway Works in the Full Time Domain

Zhaohui Zheng, Yuncheng Luo, Shaoyi Li, Zhaoyong Fan, Xi Li, Jianping Ju, Lin Ming-yu, Zijian WangSustainability 2022 · DOI

TL;DR

This paper tackles the challenge of automatically detecting construction tools on railway sites, which is crucial for safety and maintenance but difficult due to varying lighting conditions and object occlusion. They propose an enhanced YOLOX model incorporating a novel brightness enhancement module, a lightweight attention mechanism, and a CIoU loss function, achieving a significant improvement in detection accuracy and robustness across different times of day.

What Problem Does This Paper Address?

Railway maintenance and operation rely heavily on a vast array of tools and equipment. Manually tracking these items is time-consuming, prone to errors, and increasingly challenging with the expansion of high-speed rail networks. A critical issue is the need for accurate tool detection in diverse, often challenging, environments, especially during nighttime operations where visibility is poor. Existing object detection methods struggle with the wide range of lighting conditions (from bright daylight to near darkness), complex backgrounds, and the frequent overlapping or occlusion of tools, leading to missed detections and potential safety hazards. This paper aims to bridge this gap by developing a robust, full-time domain tool detection system for railway environments.

Methods

The authors build upon the YOLOX object detection framework and introduce several key improvements:

  • Brightness Enhancement Module: To address the "full time domain" requirement, they developed a novel image enhancement network. This module uses a "brightness channel decision" to categorize images into low, medium, and high brightness. Different enhancement strategies are then applied based on these categories, aiming to improve image quality without introducing excessive noise, particularly for darker images. This module is inspired by the RRDnet architecture and utilizes a three-branch convolutional neural network to decompose images into illumination, reflection, and noise components.
  • Lightweight Attention Module: To further refine feature extraction and mitigate noise introduced by the enhancement process, a lightweight attention module is integrated into the CSP layer of YOLOX. This module, inspired by CBAM, adaptively adjusts channel and spatial weights to focus on relevant features and suppress noise, improving the network's ability to handle complex scenes with overlapping objects.
  • CIoU Loss Function: To better handle the problem of densely distributed and superimposed tools, the standard YOLOX loss function is replaced with the CIoU (Complete Intersection over Union) loss. CIoU considers not only the overlap and center point distance but also the aspect ratio and shape of bounding boxes, leading to more stable and accurate bounding box regression.

Datasets

The authors created a homemade dataset of construction tools from actual railway operations, comprising 351 images. The dataset was manually labeled using the LabelImage tool, covering 16 common tool categories, including carts, generators, brooms, electric drills, cables, water pipes, woven bags, toolkits, buckets, blowers, tape measures, shovels, sand buckets, electric sanders, plastic buckets, and cement buckets. The dataset was split into training (90%) and testing (10%) sets, with 10% of the training set reserved for validation.

Key Insights and Ideas

  • Brightness is Key: The paper strongly emphasizes that lighting conditions are a primary hurdle for reliable object detection in railway environments. Their custom brightness enhancement module demonstrates that tailored image processing for different light levels is more effective than a one-size-fits-all approach.
  • Attention to Detail (Literally): The integration of a lightweight attention module highlights the power of focusing computational resources on salient features, especially when dealing with noisy or enhanced images and occluded objects.
  • Beyond Simple Overlap: The shift to CIoU loss signifies a move towards more sophisticated bounding box regression that accounts for the geometric properties of overlapping objects, crucial for distinguishing tightly packed tools.
  • Real-World Data Matters: The creation of a custom dataset from actual railway scenarios underscores the importance of domain-specific data for achieving high performance in specialized applications.

Key Results

The proposed RYOLO model achieved a mean Average Precision (mAP) of 77.26% for multiple tools, which is a 3.16% improvement over the original YOLOX model. Notably, the detection accuracy for challenging items like woven bags (often highly overlapped) saw a significant increase from an AP of 0.15 to 0.57. The model also maintained a respectable frame rate of 32.25 FPS, indicating a good balance between accuracy and speed. Qualitative results showed superior performance in low-light conditions and better handling of overlapping objects compared to Retinanet, YOLOv5s, and YOLOX.

Limitations and Future Work

The authors acknowledge that the railway works datasets available online are limited, which led them to create their own dataset. While their dataset is representative of actual scenarios, it's relatively small (351 images).

For future work, they suggest further optimizing the railway tool dataset and refining the full-time domain classification model to continue improving detection capabilities. This could involve expanding the dataset with more diverse scenarios and potentially exploring more advanced data augmentation techniques tailored to the specific challenges of railway environments.

Summary updated

Abstract

Construction tool detection is an important link in the operation and maintenance management of professional facilities in public works. Due to the large number and types of construction equipment and the complex and changeable construction environment, manual checking and inventory are still required. It is very challenging to count the variety of tools in a full-time environment automatically. To solve this problem, this paper aims to develop a full-time domain target detection system based on a deep learning network for difficult, complex railway environment image recognition. First, for the different time domain images, the image enhancement network with brightness channel decision is used to set different processing weights according to the images in different time domains to ensure the robustness of image enhancement in the entire time domain. Then, in view of the collected complex environment and the overlapping placement of the construction tools, a lightweight attention module is added on the basis of YOLOX, which makes the detection more purposeful, and the features cover more parts of the object to be recognized to improve the model. Overall detection performance. At the same time, the CIOU loss function is used to consider the distance fully, overlap rate, and penalty between the two detection frames, which is reflected in the final detection results, which can bring more stable target frame regression and further improve the recognition accuracy of the model. Experiments on the railway engineering dataset show that our RYOLO achieves a mAP of 77.26% for multiple tools and a count frame rate of 32.25FPS. Compared with YOLOX, mAP increased by 3.16%, especially the AP of woven bags with a high overlap rate increased from 0.15 to 0.57. Therefore, the target detection system proposed in this paper has better environmental adaptability and higher detection accuracy in complex railway environments, which is of great significance to the development of railway engineering intelligence.
BibTeX
@inproceedings{zheng2022rapid,
  author = {Zhaohui Zheng and Yuncheng Luo and Shaoyi Li and Zhaoyong Fan and Xi Li and Jianping Ju and Lin Ming-yu and Zijian Wang},
  title = {{Rapid Detection of Tools of Railway Works in the Full Time Domain}},
  booktitle = {Sustainability},
  year = {2022},
  doi = {10.3390/su142013662},
}
Figure