A Dynamic Watermarking Technique for Matching Communication Addresses with Cars in a Visual Field

Basic Information

• Paper ID: 2510.11353
• Title: A Dynamic Watermarking Technique for Matching Communication Addresses with Cars in a Visual Field
• Authors: Woo-Hyun Ko, Jaewon Kim, Tzu-Hsiang Lin, Samin Moosavi, P. R. Kumar
• Classification: eess.SP (Signal Processing)
• Institution: Texas A&M University
• Paper Link: https://arxiv.org/abs/2510.11353

Abstract

This paper addresses a fundamental challenge faced by intelligent roadside units (RSUs): how to match vehicles observed in visual surveillance systems with their communication IP addresses. When an RSU needs to send emergency avoidance instructions to a specific vehicle, although all vehicles are communicating with the RSU, the RSU cannot determine which vehicle in its field of view corresponds to which IP address. The paper proposes an active solution based on dynamic watermarking techniques, originally developed for cyber-physical systems security. The method requires vehicles to superimpose small random excitation signals on their steering angle or throttle/brake actuator commands and transmit this random waveform along with the IP address to the RSU. The RSU performs signal processing on the vehicle video stream to verify whether it matches the waveform in the data packet, thereby associating the packet's IP address with the corresponding vehicle in its field of view.

Research Background and Motivation

Problem Definition

The core challenge faced by intelligent transportation infrastructure systems is the matching problem between network-layer information and physical-layer data. Specifically:

1. Safety Scenario Requirements: When an RSU detects through its visual system that a vehicle is about to encounter danger, it needs to immediately send avoidance recommendations to that vehicle
2. Communication Address Matching Dilemma: The RSU knows the list of all communication IP addresses and can observe all vehicle positions, but cannot determine which IP corresponds to which vehicle
3. Limitations of Existing Solutions:
   • GPS coordinate broadcasting schemes suffer from accuracy errors and are unreliable in dense traffic scenarios
   • Received Signal Strength Indicator (RSSI)-based methods are unstable in complex environments
   • Explicit location information transmission poses privacy leakage risks

Research Significance

This problem is fundamental to vehicle-to-infrastructure (V2I) safety applications, directly affecting:

• Effectiveness of emergency avoidance systems
• Precision of intelligent traffic management
• Reliability of inter-vehicle cooperative driving

Core Contributions

1. Innovative Solution: First extension of dynamic watermarking techniques from cyber-physical systems security to vehicle identification problems
2. Privacy Protection Features: Requires only transmission of control input perturbation information without exposing specific location data
3. Real-time Matching Algorithm: Designs a real-time vehicle identification algorithm based on statistical hypothesis testing
4. Dual-stage Experimental Validation:
   • Laboratory-scale automated vehicle testing platform verification
   • Full-scale vehicle field testing in real-world road environments

Methodology Details

Dynamic Watermarking Technique Principles

Fundamental Mathematical Model

Consider a first-order stochastic linear dynamic system:

y[t+1] = ay[t] + bu[t] + w[t]

Where:

• y ∈ ℝ: System output
• u ∈ ℝ: Control input
• a, b ∈ ℝ: Known parameters
• w ~ N(0, σ²w): System noise

Watermark Injection Mechanism

The total control input is:

u[t] = u°[t] + e[t]

Where:

• u°[t]: Policy-specified control input
• e[t] ~ N(0, σ²e): Private watermark signal

Detection Mechanism

The system performs two statistical hypothesis tests:

Test 1:

lim(T→∞) (1/T)∑(k=1 to T) (z[k] - az[k-1] - bu°[k-1] - be[k-1])² = σ²w

Test 2:

lim(T→∞) (1/T)∑(k=1 to T) (z[k] - az[k-1] - bu°[k-1])² = b²σ²e + σ²w

Vehicle Identification Algorithm Design

Vehicle Motion Model

The kinematic model for the i-th vehicle:

x_i[t+1] = x_i[t] + Δt·cos(θ_i[t])·(v_i[t] + w_i,v[t])
y_i[t+1] = y_i[t] + Δt·sin(θ_i[t])·(v_i[t] + w_i,v[t])
θ_i[t+1] = θ_i[t] + Δt·(ω_i[t] + w_i,ω[t])

Where the control inputs include watermarks:

v_i[t+1] = u°_i,v(z_x, z_y, z_θ) + e_i,v[t] + w_i,v[t]
ω_i[t+1] = u°_i,ω(z_x, z_y, z_θ) + e_i,ω[t] + w_i,ω[t]

Matching Algorithm

The RSU performs matching by minimizing test statistics:

(i*, n*) = argmin(i,n∈{1,...,N}) lim(t→∞) (1/t)∑(k=1 to t) (V¹_v^(i,n)[t])²
(i*, n*) = argmin(i,n∈{1,...,N}) lim(t→∞) (1/t)∑(k=1 to t) (V²_v^(i,n)[t])²

Where:

• V¹_v^(i,n)[t] := o_n,v[t] - u°_i,v[t-1] - e_i,v[t-1]
• V²_v^(i,n)[t] := o_n,v[t] - u°_i,v[t-1]

Experimental Setup

Laboratory Testing Platform

Hardware Configuration

• Miniature Vehicles: Remote-controlled scale vehicles as test platforms
• Visual System: 10 Vicon Boneta 10 cameras with 50ms update frequency
• Control System: Model Predictive Control (MPC) algorithm
• Network Environment: Vehicular Ad-hoc Network (VANET) simulation

Experimental Scenarios

• Two vehicles traveling along elliptical trajectories
• Watermark variance settings: σ²_e1 = 0.07, σ²_e2 = 0.38
• Comparative testing of four matching scenarios

Field Testing Environment

Test Configuration

• Test Location: Texas A&M University RELLIS Campus Track
• Test Vehicles:
  • Autonomous vehicle: Lincoln MKZ equipped with drive-by-wire system
  • Manually driven vehicle: Following the autonomous vehicle
• Sensing System: Smart Micro Radar Type 48 commercial traffic monitoring radar
• Data Update Rate: 100-millisecond intervals

Experimental Parameters

• Target Speed: 10 m/s
• Watermark Variance: 0.05
• Following Distance: Approximately 2 meters
• Moving Window Average: 20 samples

Experimental Results

Laboratory Testing Results

The experiment validated four matching scenarios:

Key Findings:

• Correct matching cases 1 and 4 showed significantly lower statistical values in both tests
• Incorrect matching cases 2 and 3 displayed higher test statistics, indicating lack of correlation

Field Testing Results

Speed Change Analysis

• Autonomous Vehicle: Target speed 10 m/s with minor fluctuations due to watermark perturbations
• Manually Driven Vehicle: Relatively smooth speed variations while following

Watermark Detection Performance

• Case 1 (Autonomous Vehicle): Test statistics significantly lower than Case 2
• Case 2 (Manually Driven Vehicle): Higher test statistics indicating no correlation with watermark signal
• Moving Window Average: Effectively smoothed instantaneous fluctuations and improved detection stability

Experimental Conclusion: Field testing successfully validated the effectiveness of dynamic watermarking techniques in real-world traffic environments.

Related Work

Traditional Matching Methods

1. Signal Strength-Based Methods:
   • EV-Linker: Exploits consistency between RSSI signal changes and movement distance
   • IdentityLink: Combines camera and RF signal prediction models
   • Limitations: Signal strength unreliable in complex environments
2. GPS-Based Methods:
   • Vehicles broadcast GPS coordinates for matching
   • Limitations: GPS accuracy errors and privacy leakage risks
3. Distributed Sensing Methods:
   • RoadMap, ForeSight: Based on vehicle-mounted camera local maps
   • Limitations: Requires explicit transmission of location and color information

Dynamic Watermarking Technique Development

Dynamic watermarking techniques were originally developed for cyber-physical systems security:

• Vehicle system security protection
• Chemical process control systems
• Power grid security
• Innovation in This Paper: First application to vehicle identification problems

Conclusions and Discussion

Main Conclusions

1. Technical Feasibility: Dynamic watermarking techniques can effectively solve the RSU vehicle identification problem
2. Privacy Protection: No need to transmit sensitive location information, only control input perturbation data
3. Real-time Performance: Algorithm operates stably in real-time environments
4. Environmental Adaptability: Validated effectiveness in both laboratory and real-world road environments

Limitations

1. Watermark Amplitude Trade-off: Watermark signals must be large enough for detection but small enough to not affect normal vehicle operation
2. Multi-vehicle Scalability: Computational complexity of matching algorithm increases with vehicle count
3. Environmental Noise Impact: Strong noise environments may affect watermark detection accuracy
4. Malicious Attack Protection: Requires consideration of malicious vehicles forging watermark signals

Future Directions

1. Algorithm Optimization: Develop more efficient multi-vehicle matching algorithms
2. Adaptive Watermarking: Dynamically adjust watermark parameters based on environmental conditions
3. Security Enhancement: Incorporate cryptographic mechanisms to prevent watermark forgery
4. Standardization Application: Promote standardized deployment of the technology in practical V2I systems

In-Depth Evaluation

Strengths

1. Strong Innovation: Creatively applies dynamic watermarking techniques from network security to transportation systems
2. Clear Problem Definition: Accurately identifies and addresses fundamental problems in V2I systems
3. Solid Theoretical Foundation: Based on rigorous statistical hypothesis testing theory with complete mathematical derivations
4. Comprehensive Experimental Validation: Complete verification chain from laboratory to field testing
5. Privacy-Friendly: Avoids privacy leakage issues of traditional approaches

Weaknesses

1. Insufficient Scalability Analysis: Lacks performance analysis in large-scale multi-vehicle scenarios
2. Limited Real-time Performance Evaluation: Insufficient analysis of algorithm computational complexity and latency characteristics
3. Limited Robustness Testing: Insufficient testing for extreme weather conditions and equipment failures
4. Insufficient Security Considerations: Limited discussion of protection mechanisms against malicious attacks and watermark forgery

Impact

1. Academic Value: Provides new solution approaches for V2I system identity authentication problems
2. Practical Value: Can be directly applied to enhance intelligent transportation system security
3. Technology Promotion: Likely to promote application of dynamic watermarking techniques in more CPS domains
4. Standardization Potential: Can provide technical reference for related industry standard development

Applicable Scenarios

1. Highway RSU Systems: Particularly suitable for scenarios requiring emergency avoidance
2. Intelligent Intersections: Can be used for traffic signal coordination and conflict avoidance
3. Autonomous Vehicle Testing: Provides identity authentication means for autonomous vehicle testing
4. Fleet Management: Can be used for real-time monitoring and dispatch of commercial vehicle fleets

Overall Assessment: This is a high-quality technical paper that innovatively addresses fundamental problems in intelligent transportation systems. While there is room for improvement in scalability and security aspects, its core ideas and technical solutions possess significant academic value and practical significance.