Joint interpretation of on-board vision and static GPS cartography for determination of correct speed limit

Alexandre Bargeton, Fabien Moutarde, Fawzi Nashashibi, Anne-Sophie PuthonarXiv 2010 · arXiv

Abstract

Determining the legal speed limit is a fundamental challenge for automated driving systems. While GPS-based maps provide a baseline, they often lack real-time updates for roadworks or variable signs. Conversely, visual Traffic Sign Recognition (TSR) can be misled by signs intended for adjacent lanes or exit ramps. This paper introduces a "Speed Limit Support" system that fuses vision and GPS data using a logic-based state machine. By incorporating lane-change detection as a contextual "third sensor," the system successfully navigates complex road geometries where traditional probabilistic fusion models often fail.

Introduction

The accuracy of speed limit information is paramount for safety and regulatory compliance in modern vehicles. Existing solutions typically rely on either static GPS maps or visual recognition. However, both have significant failure modes: maps are frequently outdated, and vision systems can "over-detect" signs that do not apply to the vehicle's current path. Previous research has attempted to resolve these discrepancies using probabilistic models like Dempster-Shafer belief theory. This paper argues that such models are vulnerable to "correlated errors"—situations where both sensors are wrong in the same way—and proposes a deterministic, logic-based alternative.

Methodology

The proposed system architecture consists of three primary modules:

1. Traffic Sign Recognition (TSR)

The vision module detects circular speed signs using shape-based analysis on grayscale images. This approach ensures robustness under varying lighting conditions and at night, where color information is unreliable. Crucially, the system is designed to recognize "supplementary sub-signs," such as arrows indicating an exit ramp or vehicle-type icons, which modify the primary sign's context.

Detection of a speed limit sign with a supplementary arrow sub-sign.

2. Lane-Change Detection

To provide environmental context, the system employs a Hough transform on gradient images to track road markings. By monitoring the lateral evolution of these lines, the system acts as a "third sensor" to confirm whether a vehicle has transitioned onto an exit ramp or remained on the main highway.

3. Logic-Based Joint Interpretation

Instead of averaging sensor probabilities, the system utilizes a state machine to manage the "current" speed limit. The logic handles specific triggers:

  • Standard State: Vision-detected limits are accepted unless a sub-sign is present.
  • Exit-lane Detection: Triggered by a sign with an exit arrow.
  • Confirmation: If the lane-change detector confirms the exit, the system switches to the exit-lane speed limit and ignores highway updates.

Results and Discussion

The visual TSR system achieved a 95% accuracy rate for main signs and 80% for sub-signs. The primary advantage of the logic-based approach is its ability to handle contradictory data. In scenarios where a GPS map might suggest a highway speed (e.g., 80 km/h) while the vehicle is actually on a lower-speed exit ramp, the system uses the detected sub-sign and lane position to maintain the correct lower limit.

Joint interpretation resolving a conflict between map data (80 km/h) and visual context (50 km/h) on an exit ramp.

As shown in the experimental results, the system avoids the "correlated error" trap. If both the map and a misread sign on the main highway suggest a higher speed, the logic-based state machine prevents an incorrect update if the vehicle's trajectory (confirmed by lane detection) indicates it is on a restricted ramp. This contextual awareness allows for a more reliable output than simple probabilistic fusion, which would likely output the incorrect higher speed with high confidence.

Conclusion

By moving away from purely probabilistic fusion and toward a context-aware logic model, the authors demonstrate a more robust method for speed limit determination. Future work will focus on expanding the library of recognized sub-signs and refining the state machine to handle more complex urban road layouts, such as signs placed between closely adjacent roads.

Summary updated

Abstract

We present here a first prototype of a "Speed Limit Support" Advance Driving Assistance System (ADAS) producing permanent reliable information on the current speed limit applicable to the vehicle. Such a module can be used either for information of the driver, or could even serve for automatic setting of the maximum speed of a smart Adaptive Cruise Control (ACC). Our system is based on a joint interpretation of cartographic information (for static reference information) with on-board vision, used for traffic sign detection and recognition (including supplementary sub-signs) and visual road lines localization (for detection of lane changes). The visual traffic sign detection part is quite robust (90% global correct detection and recognition for main speed signs, and 80% for exit-lane sub-signs detection). Our approach for joint interpretation with cartography is original, and logic-based rather than probability-based, which allows correct behaviour even in cases, which do happen, when both vision and cartography may provide the same erroneous information.
BibTeX
@misc{bargeton2010joint,
  author = {Alexandre Bargeton and Fabien Moutarde and Fawzi Nashashibi and Anne-Sophie Puthon},
  title = {{Joint interpretation of on-board vision and static GPS cartography for determination of correct speed limit}},
  year = {2010},
  eprint = {1010.3867},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1010.3867},
}
Figure