AI Tracker - Multi-Target Tracking System

A state-of-the-art multi-target tracking system for aircraft tracking in clutter using Interacting Multiple Model (IMM) and Joint Probabilistic Data Association (JPDA) algorithms.

Architecture

Core Components

1. State Estimation: Kalman filters for different motion models (Constant Velocity, Constant Acceleration, Coordinated Turn)
2. IMM (Interacting Multiple Model): Manages multiple motion models and switches between them probabilistically
3. JPDA (Joint Probabilistic Data Association): Associates measurements to tracks in clutter environments
4. Simulation: Generates aircraft trajectories and clutter measurements
5. Visualization: 3D real-time visualization using egui

Algorithm Overview

IMM (Interacting Multiple Model)

The IMM algorithm maintains multiple Kalman filters running in parallel, each representing a different motion model. It:

• Mixes model-conditioned estimates based on model transition probabilities
• Updates each filter independently
• Computes model probabilities based on measurement likelihoods
• Combines estimates weighted by model probabilities

JPDA (Joint Probabilistic Data Association)

JPDA handles data association in clutter by:

• Computing association probabilities for all measurement-track pairs
• Considering all possible associations simultaneously
• Updating tracks with weighted combinations of measurements
• Handling missed detections and false alarms

Implementation Status

Implemented

• ✅ Kalman filter for Constant Velocity (CV) model
• ✅ Kalman filter for Constant Acceleration (CA) model
• ✅ Kalman filter for Coordinated Turn (CT) model
• ✅ IMM algorithm with model switching
• ✅ JPDA data association
• ✅ Aircraft simulation with realistic trajectories
• ✅ Clutter generation (Poisson-distributed false alarms)
• ✅ 3D visualization with egui
• ✅ Comprehensive unit tests

Simplified/Not Implemented

• ⚠️ Sensor model: Simplified to direct position measurements (no range/azimuth conversion)
• ⚠️ Detection probability: Constant Pd (not range-dependent)
• ⚠️ Clutter density: Uniform spatial distribution (not realistic for radar)
• ⚠️ Track initialization: Simple nearest-neighbor initialization (not M/N logic)
• ⚠️ Track deletion: Simple deletion based on missed detections (not full track quality metrics)
• ⚠️ Multiple targets: Currently optimized for single primary target tracking
• ⚠️ Sensor fusion: Single sensor only (no multi-sensor fusion)
• ⚠️ Gating: Simple ellipsoidal gating (not optimized for computational efficiency)
• ⚠️ Model parameters: Fixed transition probabilities and model parameters (not adaptive)

Building and Running

# Build the project
cargo build --release

# Run the application
cargo run --release

# Run tests
cargo test

# Run tests with output
cargo test -- --nocapture

Usage

The application provides a 3D visualization window where you can:

• View aircraft trajectories (true and estimated)
• See clutter measurements
• Observe track estimates
• Control simulation speed and parameters

Testing

All components have comprehensive unit tests. Run with:

cargo test

Test coverage includes:

• Kalman filter prediction and update
• IMM mixing and model probability updates
• JPDA association probability computation
• Measurement generation and clutter simulation