Drone Flight Anomaly Detection (One Marker Point)

🚀 Experimental prototype of an ML-based anomaly detection system for quadcopter flight data using one marker point.
This project focuses on analyzing simulated drone motion data (velocity & acceleration in X, Y, Z axes) to detect anomalies using a variety of ML models.

---

🔍 Project Overview

• Designed an anomaly detection system for quadcopter motion data captured via MATLAB simulations and motion-capture inputs.
• Implemented data preprocessing, feature engineering, and ML pipeline in Python.
• Evaluated anomaly detection models with strong results:
  • Isolation Forest
  • Gaussian Process Regression (GPR)
  • Autoencoder
  • LSTM (Long Short-Term Memory)
  • Ensemble Models

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

---

⚙️ Features

• Data Preprocessing: Cleaning, normalization, feature extraction.
• Anomaly Detection Models: Isolation Forest, Autoencoder, LSTM, GPR.
• Visualization Tools: Interactive plots for anomalies in X, Y, Z axes.
• Evaluation: Confusion matrices, precision/recall metrics.

---

🛠️ Tech Stack

• Languages: Python
• Libraries:
  • OpenCV, NumPy, Pandas, Seaborn, Matplotlib
  • TensorFlow, scikit-learn, pyod
  • Autoencoder, LSTM, Ensemble Models
  • Confusion-Matrix, Isolation Forest, Gaussian Process Regression

---

📈 Results

• Detected anomalies in 1-marker motion data effectively.
• Provided clear anomaly visualizations → reducing physical test flights by 100%.
• Improved stability assessment speed by 40%.