Intelligent system for monitoring locomotive driver alertness and actions

Volodin, Anatoliy; Sychugov, Anton; Urasinov, Daniil; Denisenko, Pavel; Andreev, Kirill; Voloschuk, Vadim · 2024 · Crossref

DOI: 10.20295/2223-9987-2024-02-86-99

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical safety issue in railway transport where human error, specifically driver fatigue and distraction, contributes to up to 70% of accidents. The authors identify a significant gap in existing locomotive monitoring systems (such as TSKBM, RPL, and CLUB), noting that they lack the capability to detect the use of mobile phones, a primary modern distraction factor. The study presents the development of an intelligent computer vision system designed to monitor locomotive drivers in real-time, aiming to enhance safety by detecting fatigue, distraction, and the presence of foreign objects like mobile phones. The system architecture integrates a video camera installed above the driver’s console with a data processing block containing neural networks. The methodology involves collecting and annotating video data of drivers in various states (awake, tired, distracted, focused) alongside physiological and behavioral data. The authors employed convolutional neural networks (CNNs) trained via supervised learning, utilizing stochastic gradient descent and backpropagation. Specific components include a PoseNetDetector using TensorFlow Lite for pose estimation and skeleton visualization, and an EmoNet model for emotion recognition based on facial features. Data augmentation techniques, including random flipping, rotation, blurring, and Gaussian noise, were applied to improve model robustness. The final classification stage utilized ensemble methods, including XGBoost, CatBoost, and Random Forest, to classify driver states based on extracted features such as facial expressions, gaze tracking, and vehicle telemetry (speed, acceleration). Experimental results indicate that the developed system successfully classifies driver states with high accuracy. Among the tested classifiers, XGBoost and Random Forest demonstrated the highest performance, with accuracy scores exceeding 0.76. The system generates reports in CSV or SQL formats, detailing the driver’s psycho-emotional state, distraction levels, and presence of objects during trips. The paper highlights that the system can detect specific events, such as a driver holding a phone, which existing systems fail to identify. The significance of this work lies in the potential integration of this vision-based monitoring system into broader railway safety complexes. By providing real-time alerts for fatigue and distraction, the system offers a proactive mechanism to prevent accidents caused by human inattention. The authors conclude that this approach allows for rapid response to emerging risks during train operation, thereby elevating the overall safety level of railway transport. The system is designed for implementation on locomotives and motor-car rolling stock to prevent emergency situations.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms