Real-time Dangerous Driving Behavior Detection using a Scalable Data Pipeline

Abstract

This project designs and implements a real-time data pipeline for detecting dangerous driving behaviors using vehicle sensor data. The pipeline ingests data from various sources, processes it using machine learning models, and delivers alerts in near real-time. The objective is to improve road safety by providing timely warnings of risky driving actions. The system employs a multi-stage architecture for efficient processing and scalability, addressing limitations of existing systems by incorporating advanced anomaly detection and edge computing. The results demonstrate improved accuracy and faster response times compared to current methods.

Introduction

Dangerous driving is a leading cause of road accidents globally, resulting in significant loss of life and property. Current methods for addressing this, like manual monitoring and limited in-vehicle warning systems, are insufficient to tackle the sheer volume of vehicles and the complexity of driving scenarios. Developing a robust, scalable, and real-time system for detecting dangerous driving behaviors is crucial. This project focuses on overcoming existing limitations by creating a sophisticated data pipeline that can handle high-volume data streams from diverse sources, enabling proactive interventions and improved road safety. The major challenge lies in developing accurate and reliable models that can handle noisy and incomplete data while minimizing false positives.

Technical Features

• Real-time streaming pipeline with Apache Kafka and Spark
• On-vehicle sensor data processing (speed, acceleration, GPS, gyroscope)
• Edge computing model for latency-sensitive inference
• Machine learning models for anomaly detection (e.g., Isolation Forest, LSTM)
• Dashboard for visualizing driving behavior and alerts