Vehicle Trajectory Tracking with Kalman Filtering Methods

This repository presents a MATLAB-based research project on vehicle trajectory tracking using classical model-based and modern learning-based Kalman filtering techniques.

Project Overview

A realistic driving scenario is generated using MATLAB’s drivingScenario framework. From this scenario, noisy position measurements are extracted and processed using three different tracking approaches:

1. Traditional IMM Kalman Filter (CV / CA)
2. Trajectory-Based Tracking (Ground-Truth Driven)
3. Unsupervised Learning-Based Adaptive Kalman Filter (KF-CA + RNN)

Traditional IMM (CV / CA)

A classical Interacting Multiple Model (IMM) Kalman filter using two motion models:

• CV (Constant Velocity)
• CA (Constant Acceleration)

Trajectory-Based Tracking

• Uses the known trajectory structure generated by the driving scenario
• Serves as a reference for visualization and benchmarking

Unsupervised Adaptive KF

A learning-based approach that does not require ground-truth states for training.

Key features:

• Constant Acceleration (CA) Kalman Filter
• RNN learns:
  • Process noise intensity
  • Measurement noise scaling

RNN inputs:

• Normalized innovation
• Normalized measurement
• Normalized time step

Evaluation Metrics

• RMSE – Root Mean Square Error
• MAE – Mean Absolute Error
• Maximum Error
• Bias per axis
• Mean NIS (Normalized Innovation Squared, consistency check)
• Rejected updates using NIS-based gating