Autonomous Terrain Relative Navigation (TRN) System

End-to-end computer vision pipeline for autonomous spacecraft descent — detects lunar/planetary surface craters via a fine-tuned YOLO model and dynamically evaluates safe landing zones through multi-factor topographic scoring.

---

Pipeline Architecture

                [Raw Descent Frame]
                           │
                           ▼
                [YOLOv8 Crater Detection]
                           │
        ┌──────────────────┴──────────────────┐
        ▼                                     ▼
[Spatial Crater Analysis]         [Terrain Hazard Modeling]
        │                                     │
        ▼                                     ▼
[Distance & Geometry Graph]     [Landing Safety Heatmap]
        │                                     │
        └──────────────────┬──────────────────┘
                           ▼
             [Coarse-to-Fine Landing Search]
                           │
                           ▼
            [Optimal Landing Point Selection]
                           │
                           ▼
              [3D Terrain Reconstruction]

1. Neural Hazard Recognition — NeuralDetector extracts crater boundaries, centroids $(c_x, c_y)$, diameters, and confidence scores via YOLO inference with NMS post-filtering.
2. Depth Estimation — Translates 2D bounding geometry into physical hazard depth estimates: $d = 0.15 \cdot \text{diameter} + 0.02 \cdot \sqrt{\text{diameter}}$
3. Circular Density Matrix — Vectorized coordinate meshes build spatial risk fields using true circular crater footprints.
4. Adaptive Suitability Scoring — Dual-pass coarse→fine grid search scores every candidate point across five normalised terms (clearance, centrality, density, smoothness, slope) within a 40 px safety buffer.
5. Telemetry Visualisation — Produces 3D terrain reconstructions, suitability heatmaps, density maps, and triangulation graphs.

---

Visual Outputs

File	Description
terrain_3d.png	Gaussian depression surface with 3D location pin at landing point
heatmap.png	Safety score field (red = hazardous → green = optimal) with annotated landing marker
density_map.png	Circular crater footprint overlay showing hazard cluster density
distances.png	Euclidean linkage graph between top-5 confidence crater centroids
localization.png	Raw descent image with detected craters and landing point marked
report.txt	Structured metrics: landing point, score, confidence statistics

---

Repository Structure

Terrain_Navigation_NN/
├── Main.py                     # Pipeline entry point
├── requirements.txt            # Dependency manifest
├── best.pt                     # Fine-tuned YOLO weights
├── data/
│   └── TRN/
│       ├── ReferenceMap.ppm    # Reference terrain map
│       └── Scene4.ppm          # Sample descent image
└── src/
    ├── __init__.py
    ├── Crater.py               # Crater data class (centerpoint, diameter, depth, radius)
    ├── ExperimentLogger.py     # Append-mode text report writer
    ├── LandingSystem.py        # Scoring functions and coarse-to-fine optimizer
    ├── NeuralNetwork.py        # YOLO inference, NMS, depth estimation
    ├── Preprocessor.py         # Centerpoint extraction utilities
    └── TerrainNavigator.py     # Pipeline orchestrator and visualisation engine

---

Installation

Requirements: Python 3.8+

# 1. Clone the repository
git clone https://github.com/Seagull28/Terrain_Navigation_NN.git
cd Terrain_Navigation_NN

# 2. Create and activate a virtual environment
python -m venv venv

# Windows
venv\Scripts\activate

# macOS / Linux
source venv/bin/activate

# 3. Install dependencies
pip install -r requirements.txt

---

Usage

python Main.py

Outputs are saved automatically to a timestamped folder:

outputs/
└── 2026-05-17_21-14-00/
    ├── report.txt
    ├── heatmap.png
    ├── terrain_3d.png
    ├── density_map.png
    ├── distances.png
    └── localization.png

---

Input Image

Output Images

Localization.png

Distances.png

Heatmap.png

terrain_3d.png

density_map.png

Sample Output

📁 Saving outputs to: outputs/2026-05-17_21-14-00
🧠 NN detected craters: 14
🧠 NN filtered craters: 12

⚠️  DETECTION ANALYSIS
----------------------
Small Craters        : 0
Low Confidence (<0.8): 5

🎯 LANDING ANALYSIS
----------------------
Best Landing Point         : (270, 280)
Landing Score              : 2.9076
Distance from nearest rim  : 54.21 px

PERFORMANCE METRICS
----------------------
Total Craters Detected: 12
Average Confidence: 0.818
Maximum Confidence: 0.885
Minimum Confidence: 0.707

---

Landing Scoring Function

The suitability score at each candidate point is computed as:

$$\text{Score} = 5.0 \cdot \hat{d}_{min} + 3.0 \cdot C_{central} - 4.0 \cdot \hat{\rho} - 2.0 \cdot \hat{s} - 2.0 \cdot \hat{g}$$

Term	Description
$\hat{d}_{min}$	Normalised clearance from nearest crater rim
$C_{central}$	Centrality reward — peaks at image centre, 0 at corners
$\hat{\rho}$	Normalised local crater density within 80 px radius
$\hat{s}$	Normalised terrain smoothness (depth-weighted influence)
$\hat{g}$	Normalised terrain slope (finite-difference gradient magnitude)

All terms are scaled to $[0, 1]$ before weighting so no single factor dominates.

---

Dependencies

ultralytics    # YOLO object detection framework
numpy          # Vectorised grid computations
matplotlib     # 3D rendering and heatmap visualisation
pillow         # Image drawing and annotation

Install via:

pip install ultralytics numpy matplotlib pillow

---

License & Citation

License

This project is licensed under the MIT License - see the LICENSE file for details. You are free to modify, distribute, and build upon this software for academic, research, or personal applications provided the original copyright notice remains intact.

Citation

If you use this autonomous navigation pipeline, the neural crater detection framework, or the topographical safety estimation algorithms in your research or academic publications, please cite the repository as follows:

@misc{trn_navigation_nn_2026,
  author       = {Thanujha Yadav},
  title        = {Autonomous Terrain Relative Navigation (TRN) System using Neural Crater Detection},
  year         = {2026},
  publisher    = {GitHub},
  journal      = {GitHub Repository},
  howpublished = {\url{[https://github.com/Seagull28/Terrain_Navigation_NN](https://github.com/Seagull28/Terrain_Navigation_NN)}}
}