TL;DR
This paper presents a system for analyzing city bus driving behavior using real-time GNSS data fused with road information. By achieving lane-level positioning and extracting speed and acceleration profiles, the system can identify risky driving patterns like speeding and harsh braking, offering a valuable tool for improving public transportation safety and management.
What Problem Does This Paper Address?
Public transportation, particularly buses, plays a crucial role in urban mobility. However, a significant number of traffic accidents and disruptions are attributed to risky driving behaviors of bus drivers. Existing methods for monitoring these behaviors often struggle with accuracy, real-time capabilities, and integration of diverse data sources. This paper aims to address this by developing a robust system that leverages high-precision GNSS positioning and road network data to provide detailed, real-time insights into bus driving patterns, enabling proactive identification and mitigation of unsafe driving practices.
Methods
The core of the proposed system is a Differential GNSS (DGNSS) pseudorange differential positioning approach, specifically a fusion of GPS and BeiDou (BDS) signals. This method enhances positioning accuracy, aiming for lane-level precision (within ±2.3 meters) which is critical for analyzing driving behavior on specific road segments.
The system architecture involves:
- Vehicle-mounted terminal: Captures real-time GNSS data (pseudorange, carrier observation data, GPGGA, GPGPD) and other vehicle information.
- Background pseudorange differential positioning: Processes the raw GNSS data using a network of CORS (Continuously Operating Reference Stations) to correct for atmospheric and satellite clock errors, thereby improving accuracy. This includes techniques like inverse distance weight interpolation for PRC (Pseudorange Correction) generation.
- Road Network Database: Integrates GIS information and digitized road maps to provide context for the vehicle's trajectory.
- Behavior Analysis Module: This module takes the processed trajectory data (position, speed, acceleration) and links it with road information (speed limits, road types) to identify risky driving behaviors. Key metrics analyzed include:
- Speeding: Compared against road-specific speed limits.
- Rapid Starting and Braking: Detected by high acceleration/deceleration values.
- Improper Stops: Analyzing stopping duration and location relative to designated bus stops.
- Average One-Way Duration: An overall evaluation metric.
- Fatigue Driving: Assessed based on continuous driving hours.
A Kalman filter is also employed for trajectory optimization to mitigate the impact of GNSS multipath and NLOS errors.
Datasets
The study utilized real-time DGNSS traces collected from multiple bus routes in Nanjing, China. The data was gathered over a week (January 1-7, 2016) for a specific bus route (Bus No. xxx). This dataset includes GNSS signals, speed, acceleration, and road section information. The paper does not explicitly mention the creation of a new benchmark dataset but rather uses collected field data for analysis.
Key Insights and Ideas
- Lane-level positioning is crucial for granular behavior analysis: Achieving sub-meter accuracy is essential to differentiate driving behaviors on specific lanes and road segments, which is not possible with standard GPS.
- Fusion of GPS and BeiDou enhances robustness: Combining signals from multiple satellite systems improves positioning reliability, especially in challenging urban environments prone to signal blockage.
- Contextualizing driving data with road information is key: Simply measuring speed and acceleration is insufficient. Linking these metrics to road speed limits, road types, and traffic conditions provides a much richer understanding of whether a behavior is truly risky or merely a response to the environment.
- Drivers adapt to real-time conditions, but risks remain: The study suggests that drivers generally adjust their driving to match road and traffic conditions. However, this adaptation can still lead to risky behaviors like speeding during low-traffic periods or harsh braking in response to unexpected events.
- Data mining reveals temporal and spatial patterns: The analysis highlights how speeding can be more prevalent during early mornings and late nights when traffic is light, and how certain road segments (e.g., Huju Road) consistently show higher percentages of risky behaviors.
Key Results
- The DGNSS system successfully achieved lane-level positioning accuracy, enabling detailed trajectory tracking and the creation of bus road networks.
- The system could automatically extract driving behavior characteristics, including speeding and severe accelerations.
- Analysis of Bus No. xxx data over a week showed varying percentages of speeding and severe accelerations across different roads, with Huju Road exhibiting higher speeding and Hanzhong Road showing fewer speeding incidents but some high accelerations.
- Speeding was observed to be higher during early mornings and late nights and lower in the afternoon, potentially linked to traffic flow and driver fatigue.
- The study identified instances of improper stopping at bus stations and noted that some drivers may temporarily disable positioning systems to avoid speeding fines, leading to data gaps.
Limitations and Future Work
Limitations: * The study acknowledges that the absolute number of speeding or accelerations is not directly comparable across different roads due to variations in total time spent on each road. They sorted by percentage of duration to address this. * Some GPS outage conditions were observed, particularly on Huju Road, which can lead to data gaps and incomplete analysis. * The paper mentions that drivers may turn off positioning modules, which is a significant limitation for continuous monitoring.
Future Work: * Improving velocity and acceleration estimations with kinematic modes for various road conditions. * Identifying bus drivers' behavior patterns more comprehensively. * Applying evaluation methods for intelligent bus systems, such as traffic light control, bus route planning, and proactive congestion management. * Further investigation into the impact of driver fatigue and its correlation with risky behaviors.
Summary updated