UAV State Estimation in GNSS-Denied Environments

Overview

Developed a modular Visual-Inertial Odometry (VIO) framework integrated with Visual Positioning System (VPS) using a custom Error-State Extended Kalman Filter (ESKF) for UAV navigation in GNSS-denied environments. This system is designed primarily for conflict zones and areas affected by GPS jamming or anti-spoofing attacks, enabling reliable autonomous flight where satellite navigation is unavailable or compromised.

Project Context

I developed a complete sensor fusion pipeline for UAV localization without GPS dependency. The system combines high-frequency IMU data with camera-based visual updates to maintain accurate pose estimation in challenging environments such as urban canyons, indoor spaces, and areas with GPS jamming or interference.

The implementation follows the OpenVINS methodology with custom extensions for VPS integration, terrain-referenced navigation, and magnetometer-based yaw correction.

System Architecture

graph TD
    classDef input fill:#e1f5fe,stroke:#01579b,stroke-width:2px,color:#000000;
    classDef process fill:#f3e5f5,stroke:#4a148c,stroke-width:2px,color:#000000;
    classDef result fill:#e8f5e9,stroke:#1b5e20,stroke-width:2px,color:#000000;
    classDef update fill:#fff3e0,stroke:#e65100,stroke-width:2px,color:#000000;

    subgraph Sensors ["Sensor Inputs"]
        direction LR
        A(["IMU<br/>(400 Hz)"])
        B(["Camera<br/>(20 Hz)"])
        C(["Magnetometer<br/>(100 Hz)"])
        D(["Barometer<br/>(5 Hz)"])
        P(["DEM<br/>(30 M Resolution)"])
    end
    
    A --> E["IMU Preintegration<br/>(Forster et al.)"]
    E --> F["ESKF State Propagation<br/>(18-State)"]
    
    B --> G["VIO Frontend<br/>(KLT + Shi-Tomasi)"]
    G --> H["MSCKF Backend<br/>(Multi-State Constraint)"]
    
    subgraph Updates ["Measurement Updates"]
        direction LR
        I["Visual Odometry<br/>(Relative)"]
        J["VPS Update<br/>(Absolute)"]
        K["Magnetometer<br/>(Yaw)"]
        L["AGL Estimation<br/>(Baro - DEM)"]
    end
    
    H --> I
    P --> J
    C --> K
    D --> L
    P --> L
    
    F --> M["Error-State Kalman Filter<br/>(Position, Velocity, Orientation,<br/>Gyro Bias, Accel Bias, Mag Bias)"]
    I --> M
    J --> M
    K --> M
    L --> M
    
    M --> N(["State Estimate<br/>(Position, Velocity, Attitude)"])
    
    class A,B,C,D,P input;
    class E,F,G,H process;
    class I,J,K,L update;
    class M process;
    class N result;

Key Components

Error-State Kalman Filter (ESKF)

• 18-state formulation: Position (3), Velocity (3), Quaternion (4→3 error), Gyro Bias (3), Accel Bias (3), Mag Bias (3)
• Manifold-based error representation: Uses rotation vector (δθ) for minimal orientation error
• Covariance propagation with camera clone management for MSCKF

VIO Frontend (OpenVINS-style)

• Grid-based feature distribution: Ensures uniform spatial coverage
• Shi-Tomasi corner detector: Better tracking quality than ORB/FAST
• Multi-stage KLT tracking: Coarse-to-fine optical flow with quality scoring
• RANSAC-based outlier rejection: Essential matrix validation

Multi-State Constraint Kalman Filter (MSCKF)

• Camera pose cloning: Maintains sliding window of past camera poses
• DLT triangulation: Linear least squares for initial 3D point estimation
• Gauss-Newton refinement: Nonlinear optimization for accurate feature positions
• Nullspace projection: Marginalize feature states to reduce computation

IMU Preintegration

• On-Manifold preintegration following Forster et al. (TRO 2017)
• Bias-corrected deltas: ΔR, Δv, Δp with first-order Jacobian correction
• Covariance propagation: Tracks uncertainty through integration period

Visual Positioning System (VPS) Integration

• Absolute position updates: Correct accumulated drift periodically
• Adaptive innovation gating: Multi-tier acceptance based on drift time
• DEM-based altitude constraint: Uses terrain data for Z-axis correction

AGL (Above Ground Level) Estimation

• Barometer-DEM fusion: Computes AGL = MSL (Barometer) − DEM height at current position
• Height constraint update: Provides Z-axis correction to EKF for altitude drift mitigation
• Adaptive noise scaling: Adjusts measurement uncertainty based on DEM availability and terrain variation

Magnetometer Processing

• Hard-iron/soft-iron calibration: Corrects sensor distortions
• EMA filtering: Smooth yaw estimates with outlier rejection
• Gyro consistency check: Validates mag readings against gyro integration

Hardware Deployment

Technologies Used

Key References

1. OpenVINS: Geneva et al., “OpenVINS: A Research Platform for Visual-Inertial State Estimation”, IROS 2020
2. IMU Preintegration: Forster et al., “On-Manifold Preintegration for Real-Time Visual-Inertial Odometry”, TRO 2017
3. MSCKF: Mourikis & Roumeliotis, “A Multi-State Constraint Kalman Filter for Vision-aided Inertial Navigation”, ICRA 2007