AN INTELLIGENT CO-DRIVER SURVEILLANCE SYSTEM

Rothkrantz, Léon; Toma, Madalina; Popa, Mirela · 2017 · OpenAlex-citations

DOI: 10.14311/app.2017.12.0083

archive: archived pipeline: cataloged verified

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

Summary

This paper presents the development and evaluation of an intelligent co-driver surveillance system designed to enhance road safety by monitoring driver behavior and generating alerts for dangerous actions. Motivated by the prevalence of driver inattention and the transitional period before fully autonomous vehicles become widespread, the authors aim to create a system that interprets semantic driving behavior through multi-modal sensor data. The system focuses on assessing visual behavior, body posture, and interaction with vehicle controls to detect missing or incorrect actions in driving sequences, such as failing to check mirrors before lane changes. The methodology employs a hierarchical model of driver activity, progressing from instantaneous body postures to specific actions, and finally to driving intentions. The system architecture integrates data from parallel sensor systems, including a Microsoft Kinect for upper-body pose estimation, an EyeLink II gaze tracker for eye and head orientation, and physical input devices like a Logitech G27 steering wheel and pedals. These sensors are synchronized with the TORCS open-source car simulator, which was customized to simulate various driving scenarios. The reasoning module utilizes both rule-based expert systems and Bayesian probabilistic models to assess the likelihood of dangerous behaviors. The rule-based system checks for missing steps in action sequences, while the probabilistic approach helps reduce false alarms by computing the probability of future steps based on historical data. Experiments were conducted with 23 student participants who drove for two hours each on simple and curved trajectories at varying speeds. The study analyzed over 16,500 single actions and 48,230 actions within scenarios. Results indicated that the system could effectively detect frequent errors, with alert correctness rates ranging from 68% for lane switching to 94% for speed limit violations. The Kinect-based pose estimation achieved an 89% accuracy rate for upper-body detection and 58% for specific body part estimation. The system successfully distinguished between normal and dangerous behaviors by analyzing temporal patterns, such as prolonged gaze diversion or lack of hand contact with the steering wheel. The significance of this work lies in its demonstration of a robust decision-support system that can assist novice drivers and enhance safety by providing real-time feedback. By fusing data from multiple sensors and employing probabilistic reasoning, the system addresses the ambiguity and noise inherent in single-sensor approaches. The findings suggest that such intelligent co-drivers can effectively monitor driver intent and physical state, offering a practical solution for reducing accidents caused by inattention and improper driving techniques in the near future.

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 qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

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

• in vehicle coaching

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: tool software, validation psychometrics
• Theoretical Contribution: computational model