Traffic Vehicle Counting in Jam Flow Conditions Using Low-Cost and Energy-Efficient Wireless Magnetic Sensors

Xu Bao, Haijian Li, Dongwei Xu, Limin Jia, Bin Ran, Jian RongSensors 2016 · DOI

TL;DR

This paper introduces a low-cost, energy-efficient wireless magnetic sensor (TrafficMS) and two algorithms for vehicle counting. They specifically address the challenging "jam flow" traffic condition, where vehicles are close together, by proposing an improved algorithm (SA-DWVDA) that significantly outperforms a basic algorithm (DWVDA) in accuracy.

What Problem Does This Paper Address?

Accurate traffic information is crucial for intelligent transportation systems (ITSs), but acquiring this data, especially vehicle counts, is difficult in "jam flow" conditions. Jam flow, common during peak hours, is characterized by high density, low speeds, and vehicles being very close (the "tailgating effect"). This close proximity creates overlapping magnetic signals, confusing standard vehicle detection algorithms. Existing methods often struggle with this specific scenario, leading to inaccurate counts. The paper aims to provide a robust and cost-effective solution for vehicle counting in these challenging conditions.

Methods

The core of the solution is the TrafficMS, a wireless magnetic sensor designed for low cost and energy efficiency. It utilizes AMR (Anisotropic Magneto-Resistance) sensors to detect disturbances in the Earth's magnetic field caused by passing vehicles.

Two algorithms are proposed:

  • Basic Vehicle Detection Algorithm (DWVDA): This algorithm uses a double-window approach to identify vehicle approaching and departing points based on the magnetic signal. It's designed for less congested traffic where vehicles have larger headways.
  • Improved Vehicle Detection Algorithm for Jam Flow Conditions (SA-DWVDA): This algorithm builds upon the DWVDA by introducing a second adaptive mechanism. This mechanism allows the sensor to "learn" the traffic environment, specifically adapting to the baseline magnetic signal deviations caused by the "tailgating effect" in jam flow. It uses an additional window (w3) and a forgetting factor (α) to dynamically update the baseline, making it more resilient to the close-proximity vehicle signals.

Datasets

The paper describes two field experiment scenarios:

  • Scenario I: TrafficMS nodes were deployed on a sub-arterial road in Beijing, capturing data during synchronized flow and jam flow conditions.
  • Scenario II: TrafficMS nodes were deployed at a signalized intersection in Xi'an, China, covering a 24-hour period with free-flow, synchronized flow, and jam flow conditions.

Actual traffic volumes for comparison were obtained through manual statistics from video recordings. No new benchmark datasets were introduced; rather, existing traffic conditions were used for evaluation.

Key Insights and Ideas

  • The "Tailgating Effect" is the Key Challenge: The paper clearly identifies the close proximity of vehicles in jam flow as the primary obstacle for magnetic sensors, causing signal interference that degrades accuracy.
  • Adaptive Learning is Crucial for Jam Flow: A static detection algorithm struggles in jam flow. The SA-DWVDA's ability to adapt to the changing baseline magnetic field caused by tailgating vehicles is what gives it a significant advantage.
  • Low-Cost Sensors Can Be Effective: The research demonstrates that inexpensive, energy-efficient magnetic sensors can provide valuable traffic data, making widespread deployment feasible.
  • Robustness Across Conditions: While SA-DWVDA excels in jam flow, both algorithms perform well in less congested scenarios, indicating a degree of robustness.

Key Results

  • Low Traffic Volume Conditions (Scenario I): Both DWVDA and SA-DWVDA achieved satisfactory performance with Mean Absolute Percent Errors (MAPEs) less than 1%.
  • Jam Flow Conditions (Scenario II):
    • The SA-DWVDA achieved a MAPE of 2.54%, significantly outperforming the DWVDA's MAPE of 7.07%.
  • In specific jam flow datasets within Scenario II, the DWVDA's MAPE increased to 15-20%, while SA-DWVDA's MAPE remained around 2%.
  • The SA-DWVDA demonstrated improved accuracy in dealing with "tailgating effect" (TE) and "big vehicle effect" (BE) errors compared to DWVDA.

Limitations and Future Work

  • Single-Lane Focus: The experiments were primarily conducted on single lanes. The authors suggest future work should include multi-lane and multi-section scenarios to better reflect real-world complexity.
  • Project Investment and Site Limitations: The current study was limited by project investment and the number of experimental sites.
  • Algorithm Complexity: The SA-DWVDA's adaptive nature increases its computational complexity compared to DWVDA.
  • Future Directions:
    • Extend experiments to multi-lane and multi-section traffic.
  • Conduct long-term, continuous traffic flow data collection across various traffic states (freeways, ramps, urban expressways) to support large-scale WSN applications.
  • Investigate WSN-based detection and evaluation methods for multiple traffic flow parameters beyond just counting.
  • Consider a dual-algorithm approach (DWVDA for free-flow, SA-DWVDA for jam flow) or a time-based activation strategy to balance performance and computational load.

Summary updated

Abstract

The jam flow condition is one of the main traffic states in traffic flow theory and the most difficult state for sectional traffic information acquisition. Since traffic information acquisition is the basis for the application of an intelligent transportation system, research on traffic vehicle counting methods for the jam flow conditions has been worthwhile. A low-cost and energy-efficient type of multi-function wireless traffic magnetic sensor was designed and developed. Several advantages of the traffic magnetic sensor are that it is suitable for large-scale deployment and time-sustainable detection for traffic information acquisition. Based on the traffic magnetic sensor, a basic vehicle detection algorithm (DWVDA) with less computational complexity was introduced for vehicle counting in low traffic volume conditions. To improve the detection performance in jam flow conditions with a "tailgating effect" between front vehicles and rear vehicles, an improved vehicle detection algorithm (SA-DWVDA) was proposed and applied in field traffic environments. By deploying traffic magnetic sensor nodes in field traffic scenarios, two field experiments were conducted to test and verify the DWVDA and the SA-DWVDA algorithms. The experimental results have shown that both DWVDA and the SA-DWVDA algorithms yield a satisfactory performance in low traffic volume conditions (scenario I) and both of their mean absolute percent errors are less than 1% in this scenario. However, for jam flow conditions with heavy traffic volumes (scenario II), the SA-DWVDA was proven to achieve better results, and the mean absolute percent error of the SA-DWVDA is 2.54% with corresponding results of the DWVDA 7.07%. The results conclude that the proposed SA-DWVDA can implement efficient and accurate vehicle detection in jam flow conditions and can be employed in field traffic environments.
BibTeX
@inproceedings{bao2016traffic,
  author = {Xu Bao and Haijian Li and Dongwei Xu and Limin Jia and Bin Ran and Jian Rong},
  title = {{Traffic Vehicle Counting in Jam Flow Conditions Using Low-Cost and Energy-Efficient Wireless Magnetic Sensors}},
  booktitle = {Sensors},
  year = {2016},
  doi = {10.3390/s16111868},
}
Figure