Drone Flight Anomaly Detection (1 & 5 Marker Points)

🚀 Final implementation of a drone flight anomaly detection system using 1-marker and 5-marker motion-capture datasets.
This project extends the one-marker prototype into a robust system capable of detecting anomalies in multi-point drone motion data.

---

🔍 Project Overview

• Built a machine learning pipeline for anomaly detection using Python.
• Trained on both 1-marker and 5-marker motion-capture data (velocity & acceleration on X, Y, Z axes).
• Implemented anomaly detection with multiple models:
  • Isolation Forest
  • Gaussian Process Regression (GPR)
  • One-Class SVM
  • LSTM Autoencoder
  • Ensemble Models

📊 Achieved 92% detection accuracy with 5% false positive rate.

---

⚙️ Features

• Data Preprocessing: Cleaning, train-test split, normalization, feature engineering.
• Model Training: One-Class SVM, Isolation Forest, GPR, LSTM Autoencoder.
• Evaluation: Confusion matrices, ROC-AUC, precision-recall metrics.
• Visualization: Per-axis anomaly representation for stability analysis.
• Joblib Model Saving: Export trained models for reusability.

---

🛠️ Tech Stack

• Languages: Python
• Libraries:
  • TensorFlow, Keras, NumPy, Pandas, Seaborn, Matplotlib
  • scikit-learn, pyod, joblib
  • Confusion-Matrix, Isolation Forest, Gaussian Process Regression, One-Class SVM
  • LSTM Autoencoder, Test-Train Split

---

📈 Results

• Extended anomaly detection from 1-marker prototype → 5-marker full implementation.
• Anomalies represented per marker → improved accuracy & robustness.
• Reduced dependency on physical test flights by 100%.
• Increased stability assessment efficiency by 40%.