Our research focuses on developing a comprehensive drone regulation framework for low-altitude emergency unmanned aerial vehicles operating in mountainous highway spiral tunnel groups. The enclosed environment, complex terrain, and limited emergency access of these tunnel groups pose significant challenges to traditional rescue operations. Through real-vehicle experiments and multi-dimensional data analysis, we have established a risk-based drone regulation strategy that optimizes patrol frequency according to driving risk levels.
1. Introduction
The rapid expansion of highway networks into mountainous regions has led to an increasing number of tunnel groups with complex geometries, including spiral tunnels. These structures present unique challenges for emergency response due to their enclosed nature, limited ventilation, restricted access points, and challenging terrain. Traditional emergency rescue methods relying on ground-based equipment and personnel often suffer from delayed response times, information gaps, and coordination inefficiencies. In this context, drone regulation becomes critical for ensuring that unmanned aerial vehicles can effectively support emergency operations while maintaining safety and operational efficiency.
Low-altitude drones offer significant advantages for emergency response in tunnel groups, including rapid deployment, high mobility, flexible sensor payloads, and the ability to access hard-to-reach areas. However, the effective utilization of drones in this specialized environment requires robust drone regulation strategies that account for the unique operational constraints, including GPS signal degradation, complex airflow patterns, multi-drone coordination requirements, and the need to prioritize emergency missions over routine patrols.
The concept of drone regulation in transportation emergency management has evolved significantly in recent years. While previous studies have explored drone applications in individual tunnels or open highway sections, there remains a critical gap in developing integrated drone regulation strategies specifically designed for spiral tunnel groups—where multiple tunnels are connected in sequence with extreme curvature and elevation changes. Our research addresses this gap by establishing a data-driven, risk-based drone regulation framework that aligns patrol intensity with actual driving risk levels.
2. Real-Vehicle Driving Experiment
2.1 Experimental Conditions
We conducted real-vehicle driving experiments in a representative spiral tunnel group located in a mountainous region. The experimental section consisted of three consecutive spiral tunnels with the following characteristics:
| Parameter | Tunnel A | Tunnel B | Tunnel C |
|---|---|---|---|
| Length (m) | 1330 | 2100 | 2200 |
| Design Speed (km/h) | 80 | 80 | 80 |
| Clear Height (m) | 5 | 5 | 5 |
| Clear Width (m) | 10.9 | 10.9 | 10.9 |
| Maximum Radius (m) | 2200 | 2500 | 1200 |
| Minimum Radius (m) | 850 | 850 | 730 |
| Longitudinal Gradient (%) | 2.45 | — | — |
We recruited 40 licensed drivers through stratified sampling based on age, gender, occupation, and education level. The sample included 30 male and 10 female participants, with ages ranging from 20 to 70 years (mean age: 41.5 years, standard deviation: 15.6 years). To minimize individual variability effects, we implemented standardized experimental procedures, data normalization protocols, and consistent equipment calibration across all trials.
The experimental equipment comprised four categories: vehicle systems, data acquisition devices, recording equipment, and output systems. The test vehicle was an automatic transmission passenger car. On-board diagnostics (OBD) and inertial measurement units (IMU) captured vehicle dynamics, while illuminance meters and noise dosimeters recorded tunnel environmental conditions. Driver physiological data were collected using Tobii Glasses 3 eye-trackers and ErgoLAB wearable physiological recorders. Two cameras mounted on the windshield recorded the driver’s behavior and road conditions simultaneously.
2.2 Experimental Procedure
The experiment route spanned 38.5 km from the starting toll station to the ending toll station, with round trips conducted at different times of the day. The procedure consisted of preparation and formal testing phases. During preparation, participants completed questionnaires, received instructions about experimental requirements, and underwent equipment calibration and system testing. In the formal phase, all devices simultaneously collected vehicle dynamics, tunnel environmental parameters, and driver physiological data while participants drove according to their natural driving habits.
This comprehensive data collection approach enabled us to capture the full spectrum of factors influencing driving risk in spiral tunnel environments, forming the foundation for our subsequent drone regulation strategy development.
3. Driving Risk Quantification
3.1 Indicator Selection and Correlation Analysis
We initially selected nine indicators covering the driver-vehicle-road-environment dimensions: heart rate (HR), pupil area (PA), vehicle speed (v), longitudinal acceleration (a), lateral offset (d), steering wheel angle (SW), illuminance (Lx), road radius (R), and road gradient. Pearson correlation analysis revealed significant correlations among eight of these indicators, while road gradient showed no significant correlation with the others. Therefore, we excluded road gradient from further analysis.
| Indicator | HR | PA | v | a | d | SW | Lx | R |
|---|---|---|---|---|---|---|---|---|
| HR | 1.000 | 0.656*** | 0.551*** | 0.713*** | 0.535*** | 0.525*** | -0.451*** | -0.745*** |
| PA | 1.000 | 0.595*** | 0.644*** | 0.420*** | 0.476*** | -0.821*** | -0.539*** | |
| v | 1.000 | 0.611*** | 0.796*** | -0.638*** | 0.623*** | 0.780*** | ||
| a | 1.000 | 0.682*** | -0.271*** | 0.198*** | 0.621*** | |||
| d | 1.000 | 0.839*** | 0.662*** | -0.642*** | ||||
| SW | 1.000 | 0.509*** | -0.869*** | |||||
| Lx | 1.000 | 0.345*** | ||||||
| R | 1.000 |
Note: *** indicates significance at p < 0.001 level.
3.2 Factor Analysis
The Kaiser-Meyer-Olkin (KMO) test yielded a value of 0.726, indicating acceptable sampling adequacy for factor analysis. Bartlett’s test of sphericity was significant (p < 0.001), confirming the presence of sufficient correlations among indicators. Using principal component extraction with varimax rotation, we identified three factors with eigenvalues greater than 1, collectively explaining 86.06% of the total variance.
| Component | Eigenvalue | Variance (%) | Cumulative (%) |
|---|---|---|---|
| Factor 1 | 3.872 | 48.40 | 48.40 |
| Factor 2 | 2.045 | 25.56 | 73.96 |
| Factor 3 | 1.028 | 12.10 | 86.06 |
The rotated component matrix revealed the following factor structure:
- Factor 1 (Lateral Stability Factor): High loadings from lateral offset (0.831), steering wheel angle (-0.959), and road radius (-0.958), explaining 47.41% of the variance.
- Factor 2 (Physiological Load Factor): High loadings from pupil area (-0.951), illuminance (0.927), and heart rate (0.460), explaining 25.13% of the variance.
- Factor 3 (Longitudinal Control Factor): High loadings from acceleration (0.967) and speed (0.893), explaining 13.52% of the variance.
3.3 Driving Risk Quantification Model
Based on the factor analysis results, we derived the following component score functions:
For Factor 1 (Lateral Stability):
$$SF_1 = -0.151 \cdot z_{HR} – 0.077 \cdot z_{PA} + 0.007 \cdot z_v + 0.016 \cdot z_a + 0.220 \cdot z_d – 0.262 \cdot z_{SW} + 0.010 \cdot z_{Lx} – 0.256 \cdot z_R$$
For Factor 2 (Physiological Load):
$$SF_2 = 0.238 \cdot z_{HR} – 0.511 \cdot z_{PA} + 0.040 \cdot z_v + 0.073 \cdot z_a + 0.003 \cdot z_d – 0.093 \cdot z_{SW} + 0.448 \cdot z_{Lx} – 0.035 \cdot z_R$$
For Factor 3 (Longitudinal Control):
$$SF_3 = 0.224 \cdot z_{HR} – 0.140 \cdot z_{PA} + 0.239 \cdot z_v + 0.916 \cdot z_a – 0.175 \cdot z_d – 0.115 \cdot z_{SW} – 0.089 \cdot z_{Lx} – 0.122 \cdot z_R$$
Where \(z_{x}\) represents the standardized value of each indicator, calculated as:
$$z_x = \frac{x – \bar{x}}{s}$$
where \(\bar{x}\) is the mean and \(s\) is the standard deviation.
The comprehensive driving risk score (SF) is then computed as a weighted composite of the three factors:
$$SF = \frac{1}{0.8606}(0.4741 \cdot SF_1 + 0.2513 \cdot SF_2 + 0.1352 \cdot SF_3)$$
This model integrates vehicle dynamics, driver physiological responses, and environmental conditions into a single quantitative risk metric, providing a robust foundation for risk-based drone regulation.
3.4 Risk Classification and Spatial Visualization
Using the k-means++ clustering algorithm on the computed SF values, we identified four distinct risk levels with the following threshold ranges:
| Risk Level | Risk Score Range | Category Label |
|---|---|---|
| Low | [1.25, 2.61) | L |
| Medium-Low | [2.61, 4.13) | ML |
| Medium-High | [4.13, 5.50) | MH |
| High | [5.50, 7.63] | H |
The spatial distribution of risk across the three tunnels is summarized below:
| Risk Level | Tunnel A (%) | Tunnel B (%) | Tunnel C (%) |
|---|---|---|---|
| Low | 15.06 | 10.11 | 5.87 |
| Medium-Low | 51.13 | 46.13 | 46.33 |
| Medium-High | 22.54 | 32.51 | 35.09 |
| High | 11.27 | 11.25 | 12.71 |
The spatial risk visualization reveals that higher risk segments are concentrated in sections with smaller curve radii, transition zones between curves, and areas with significant illuminance variations. This spatial heterogeneity underscores the need for differentiated drone regulation strategies rather than uniform patrol approaches.
4. UAV Patrol Control Strategy Based on Risk Levels
4.1 Concept of Risk Exposure
To translate the quantified driving risk into actionable drone regulation parameters, we introduce the concept of risk exposure—borrowed from reliability engineering and risk-based inspection (RBI) methodologies. In our framework, risk exposure represents the cumulative risk accumulated over a patrol cycle and is defined as the product of the representative risk value and the patrol period.
For a given risk level \(g \in \{L, ML, MH, H\}\), the risk exposure within one patrol cycle is:
$$E_g^k = F_g^* \cdot T_g^k$$
where:
- \(F_g^*\) = representative risk score for level \(g\) (taken as the midpoint of the corresponding interval)
- \(T_g^k\) = patrol period for risk level \(g\) in tunnel \(k\) (minutes)
- \(E_g^k\) = risk exposure accumulated during one patrol cycle
The representative risk scores for each level are:
- Low (L): \(F_L^* = 1.93\)
- Medium-Low (ML): \(F_{ML}^* = 3.37\)
- Medium-High (MH): \(F_{MH}^* = 4.81\)
- High (H): \(F_H^* = 6.56\)
4.2 Calculation of Patrol Period and Frequency
To prevent excessive risk accumulation in any single patrol cycle, we impose the constraint that the risk exposure should not exceed a tunnel-specific upper limit constant \(C^k\):
$$E_g^k = C^k$$
This yields the patrol period for each risk level:
$$T_g^k = \frac{C^k}{F_g^*}$$
The upper limit constant \(C^k\) is calibrated using the weighted average baseline risk of the tunnel and a standard baseline patrol period \(T_{base} = 15\) minutes:
$$C^k = \bar{F}_w^k \cdot T_{base}$$
The weighted average baseline risk for tunnel \(k\) is:
$$\bar{F}_w^k = w_L^k \cdot F_L^* + w_{ML}^k \cdot F_{ML}^* + w_{MH}^k \cdot F_{MH}^* + w_H^k \cdot F_H^*$$
where \(w_g^k\) represents the proportion of segments at risk level \(g\) in tunnel \(k\).
Based on the proportions from Table 5, we computed the following weighted average baseline risks:
- Tunnel A: \(\bar{F}_w^A = 3.84\)
- Tunnel B: \(\bar{F}_w^B = 4.05\)
- Tunnel C: \(\bar{F}_w^C = 4.20\)
Corresponding exposure limits:
- Tunnel A: \(C^A = 57.60\)
- Tunnel B: \(C^B = 60.75\)
- Tunnel C: \(C^C = 63.00\)
Substituting these values into the patrol period formula yields:
| Risk Level | Tunnel A | Tunnel B | Tunnel C |
|---|---|---|---|
| Low | 29.84 | 31.48 | 32.64 |
| Medium-Low | 17.09 | 18.03 | 18.69 |
| Medium-High | 11.98 | 12.63 | 13.10 |
| High | 8.78 | 9.26 | 9.60 |
Converting periods to patrol frequencies (patrols per hour):
$$f_g^k = \frac{60}{T_g^k}$$
| Risk Level | Tunnel A | Tunnel B | Tunnel C |
|---|---|---|---|
| Low | 2.01 | 1.91 | 1.84 |
| Medium-Low | 3.51 | 3.33 | 3.21 |
| Medium-High | 5.01 | 4.75 | 4.58 |
| High | 6.83 | 6.48 | 6.25 |
4.3 Differentiated Patrol Strategy for Drone Regulation
Based on the calculated patrol frequencies, we formulate the following risk-based drone regulation strategy for emergency UAV operations in spiral tunnel groups:
| Risk Level | Patrol Frequency (patrols/h) | Patrol Category | Operational Priority |
|---|---|---|---|
| Low | ≈ 2 | Low-frequency patrol | Routine monitoring |
| Medium-Low | 3 – 4 | Standard patrol | Regular inspection |
| Medium-High | ≈ 5 | Enhanced patrol | Increased vigilance |
| High | ≥ 6 | High-frequency patrol | Priority monitoring |
This differentiated drone regulation strategy enables efficient allocation of limited UAV resources by matching patrol intensity to actual risk levels. The key operational principles include:
- Low-risk segments (≈2 patrols/h): Implement routine monitoring with basic sensor configurations. These segments require minimal intervention and can be patrolled with longer intervals, allowing drone resources to be reallocated to higher-risk areas.
- Medium-low risk segments (3-4 patrols/h): Conduct standard patrols with regular sensor payloads. These segments represent typical driving conditions where regular monitoring suffices for early anomaly detection.
- Medium-high risk segments (≈5 patrols/h): Deploy enhanced patrols with advanced sensors (e.g., thermal imaging, high-resolution cameras). These segments warrant increased attention due to elevated accident probability and potential severity.
- High-risk segments (≥6 patrols/h): Implement intensive patrols with full sensor suites and real-time data transmission. These segments require continuous monitoring and rapid response capabilities. drones should be pre-positioned near these segments to minimize response time in case of incidents.
The integration of drone regulation with risk quantification provides several operational benefits:
- Resource optimization: By concentrating UAV patrol efforts on high-risk segments, our strategy maximizes the utility of limited drone assets while maintaining adequate coverage across all risk levels.
- Proactive emergency response: The risk-based patrol frequency ensures that high-risk segments receive frequent monitoring, enabling early detection of incidents before they escalate.
- Adaptive management: The drone regulation framework can be updated dynamically as new data becomes available, allowing patrol strategies to evolve with changing traffic patterns, weather conditions, and infrastructure modifications.
- Safety enhancement: By maintaining appropriate patrol frequencies across all risk levels, the strategy contributes to overall traffic safety by deterring hazardous driving behaviors and enabling rapid response to emergencies.
5. Discussion
5.1 Implications for Drone Regulation in Mountainous Tunnel Groups
Our research demonstrates that effective drone regulation in spiral tunnel groups requires a paradigm shift from uniform patrol approaches to risk-based differentiated strategies. The significant spatial heterogeneity in driving risk across tunnel segments—with high-risk concentrations near sharp curves, transition zones, and areas with poor illumination—necessitates corresponding heterogeneity in patrol intensity.
The drone regulation strategy we developed provides a quantitative framework for determining appropriate patrol frequencies based on objectively measured risk levels. This approach offers several advantages over traditional fixed-interval patrol schemes:
- Evidence-based decision making: Patrol frequencies are derived from empirical data rather than subjective judgment, ensuring that resources are allocated where they are most needed.
- Scalability: The methodology can be extended to other tunnel groups by recalibrating the risk model parameters based on local conditions.
- Integration with broader traffic management systems: The risk scores can be incorporated into intelligent transportation systems for coordinated drone regulation and emergency response.
5.2 Challenges and Future Directions
While our drone regulation strategy provides a solid foundation for emergency UAV operations in tunnel groups, several challenges remain that warrant further investigation:
- Dynamic risk evolution: Our current model assumes static risk levels based on infrastructure geometry and average driving behavior. Future work should incorporate real-time traffic conditions, weather data, and incident information to enable dynamic drone regulation adjustments.
- Multi-drone coordination: The proposed strategy focuses on individual drone patrol frequencies. Future research should address coordination mechanisms for multiple drones operating simultaneously in the same tunnel group, including conflict resolution, task allocation, and communication protocols.
- GPS-denied navigation: Spiral tunnels present significant challenges for GPS-based navigation. Robust drone regulation must incorporate alternative localization methods such as visual SLAM, LiDAR-based mapping, or sensor fusion approaches to ensure reliable operation in tunnel environments.
- Regulatory compliance: The drone regulation strategy must align with evolving aviation authority requirements for beyond visual line of sight (BVLOS) operations, autonomous flight, and emergency procedures in complex airspace.
6. Conclusion
This research presents a comprehensive drone regulation framework for low-altitude emergency UAV operations in mountainous highway spiral tunnel groups. Through real-vehicle experiments, multi-dimensional data collection, and rigorous statistical analysis, we developed a driving risk quantification model that captures the combined effects of vehicle dynamics, driver physiology, and environmental conditions. The model classifies tunnel segments into four risk levels, enabling spatial visualization of risk distribution.
The key contribution of our work lies in establishing a quantitative link between driving risk and drone regulation parameters. By introducing the concept of risk exposure and calibrating exposure limits against a standard baseline patrol period, we derived risk-specific patrol frequencies that ensure proportional resource allocation across different risk segments. The resulting differentiated drone regulation strategy prescribes ≈2 patrols/h for low-risk segments, 3-4 patrols/h for medium-low risk, ≈5 patrols/h for medium-high risk, and ≥6 patrols/h for high-risk segments.
Our drone regulation strategy offers a practical, evidence-based approach to optimizing UAV patrol operations in challenging tunnel environments, enhancing both emergency response capabilities and operational efficiency. The methodology is transferable to other tunnel groups with appropriate recalibration, providing a valuable tool for transportation agencies seeking to integrate drone technology into their safety management systems. Future work will focus on incorporating dynamic risk factors, multi-drone coordination mechanisms, and advanced navigation solutions to further enhance the robustness and adaptability of drone regulation in mountainous tunnel environments.