Real time distracted driver detection using Xception architecture and Raspberry Pi

Narayanan, Uma; Prajith, Pavan; Mathew, Rijo Thomas; Alexandar, Royal; Vikraman, Vishnu · 2024 · Crossref

DOI: 10.4114/intartif.vol28iss75pp15-29

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 addresses the critical safety issue of distracted driving, which contributes significantly to global road traffic fatalities and accidents. The authors aim to develop a reliable, real-time detection system that identifies distracted drivers and notifies them of their lack of focus. The motivation stems from the limitations of traditional reactive safety measures and the need for proactive, deep learning-based interventions. The study specifically seeks to balance three often conflicting requirements for real-time edge deployment: high accuracy, fast processing speed, and a small number of model parameters. The proposed solution utilizes Convolutional Neural Networks (CNNs) to classify driver behaviors using the State Farm Distracted Driver Dataset (SFDDD), which contains images of ten distinct driving activities, such as texting, calling, and normal driving. The experimental design involved preprocessing the dataset and training three specific CNN architectures: ResNet50, VGG16, and Xception. The models were trained using transfer learning on the ImageNet dataset, with custom fully connected layers added for classification. The training was conducted on Google Colab with GPU acceleration, splitting the data into 80% for training and 20% for testing. For real-time implementation, the system integrates a Raspberry Pi Model 3B+ with a 16 MP Arducam camera to capture driver images and trigger an alarm upon detecting distraction. The results indicate that while ResNet50 achieved an accuracy of 88% and VGG16 reached 86.89%, both models produced inaccurate predictions on specific test cases, likely due to overfitting. The Xception model achieved a final accuracy of 85.58%. Despite having a lower accuracy metric than the other two models, Xception demonstrated superior performance in correctly predicting test cases and maintaining stability. The authors concluded that Xception was the optimal choice for this application because it avoided the overfitting issues observed in ResNet50 and VGG16, providing more reliable real-time detection. The significance of this work lies in its demonstration of a viable, low-cost hardware-software integration for proactive driver monitoring. By selecting Xception, the authors prioritized robust generalization and real-time applicability over raw training accuracy, addressing the practical constraints of deploying deep learning models on edge devices like the Raspberry Pi. This approach offers a scalable method for reducing accident risks through immediate alerts, contributing to the broader field of intelligent transportation systems and autonomous vehicle safety protocols.

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.

• distraction detection algorithms
• visual

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