Skip to content

Comprehensive Tracker Benchmarking, Selection & Strategy Guide for OpenPTV2

Executive Summary & Key Findings

OpenPTV2 features five 3D tracking engines, each engineered for distinct flow regimes, particle densities, and execution constraints. Based on empirical cross-tracker benchmarking on turbulent synthetic flow fields, there is no single "best" tracker for every flow, but rather an optimal tracker or hybrid pipeline depending on dataset characteristics.

1. Single-Pass Engine Benchmarks

Evaluated on turbulent flow dataset (Mean $N \approx 220$ particles/frame, $v = 1.52\text{ mm/frame}, d_{\text{nn}} = 6.98\text{ mm}, M = 0.218$):

Tracker Engine Precision Recall / Yield Fragment Count ($F$) Track Purity Perfect Match % Speed (ms/frame)
priority_segment_3d (Default) 0.974 0.901 3.56 0.970 73.7% 159.4 ms
nearest_hungarian_3d 0.984 0.904 3.53 0.982 76.2% 150.6 ms
predictive_gmm_3d 0.959 0.752 7.53 0.977 86.0% 732.3 ms

2. Hybrid Cascading Strategies Benchmark

To go beyond single-pass limits, OpenPTV2 supports multi-pass hybrid cascading workflows (scripts/demo_hybrid_strategies.py):

Hybrid Strategy Pipeline Precision Recall / Yield Ghost % Fragment Count ($F$) Track Purity Perfect Match % (PMT) Speed (ms/frame)
Baseline (Single Pass priority_segment_3d) 0.974 0.901 3.56% 3.56 0.970 73.7% 157.0 ms
Strategy 1: Forward-Fast / Backward-GMM 0.974 0.901 0.50% 3.15 0.965 93.6% 689.9 ms
Strategy 2: Two-Scale Velocity Cascading 0.963 0.888 1.07% 3.67 0.938 87.0% 365.4 ms

Strategy Highlights

  • Strategy 1 (Forward Fast / Backward GMM): Increases Perfect Match Trajectory Percentage (PMT%) from 73.7% to 93.6% and reduces Ghost capture rate from 3.56% down to 0.50%.
  • Strategy 2 (Two-Scale Velocity Cascading): Achieves 87.0% PMT and reduces Ghost capture to 1.07% at nearly double the speed of Strategy 1 (365.4 ms/frame).

3. Tracker Selection Decision Matrix

                          [ What is your Flow Regime & Density? ]
                                            │
           ┌────────────────────────────────┴────────────────────────────────┐
           ▼                                                                 ▼
[ High / Medium Density ]                                         [ Low Density / Sparse ]
    (N > 500 / frame)                                                (N < 200 / frame)
           │                                                                 │
   ┌───────┴────────┐                                                        │
   ▼                ▼                                                        ▼
[ Smooth Flow ]  [ Strong Turbulence ]                             [ High Noise / Missing ]
   │                │                                                        │
   ▼                ▼                                                        ▼
priority_segment_3d nearest_hungarian_3d                             predictive_gmm_3d

Condition-by-Condition Guide

Flow Condition Recommended Engine / Hybrid Primary Reason
High Density / Large Datasets ($N > 1,000$) priority_segment_3d $O(N)$ spatial grid cell indexing provides 4x–10x faster runtime without precision loss.
High Velocity Turbulences & Recirculation nearest_hungarian_3d Global Hungarian bipartite matching prevents greedy assignment conflicts in dense eddies.
Irregular Steps / High Noise predictive_gmm_3d GMM-based predictive estimation handles complex trajectory curvatures and noise.
Fragmented Tracks needing High Purity Strategy 1 Hybrid 93.6% PMT with 0.50% Ghost Capture rate.

4. Parameter Tuning Reference

A. priority_segment_3d (Cython Engine)

  • dacc (Maximum Acceleration Window, mm):
  • Default: 5.5 mm
  • Tuning: Do not expand dacc arbitrarily! Expanding dacc from 5.5 to 50 drops precision from 0.974 to 0.810 due to enlarged candidate search bounds.
  • angle (Maximum Turn Angle, deg):
  • Default: 120.0 deg
  • Tuning: Decrease to 45.0–60.0 deg for directed laminar flows to prune improbable sharp turns.

B. nearest_hungarian_3d (MyPTV Kinematic Engine)

  • search_radius (Max Displacement Distance, mm):
  • Default: Derived from dvxmax (15.5 mm).
  • Tuning: Set search radius to $1.5 \times v_{\max} \Delta t$. Setting radius too large ($>3 \times v_{\max}$) degrades precision in dense regions.

5. Cheat Sheet & Execution Commands

Target Objective Recommended Configuration / Strategy Command
Default Fast Pass Single Pass priority_segment_3d uv run python scripts/bench_trackers.py
Maximum Trajectory Perfection (93.6% PMT) Hybrid Strategy 1 (Forward Fast / Backward Kalman) uv run python scripts/demo_hybrid_strategies.py
Fast Multi-Scale Flow (87.0% PMT) Hybrid Strategy 2 (Two-Scale Cascading) uv run python scripts/demo_hybrid_strategies.py
High Density / Large Scale Spatial Grid Accelerated priority_segment_3d uv run python scripts/bench_trackers.py --density 5000

6. Future Verification & Benchmarking TODO Backlog

To continuously validate and expand OpenPTV2 tracking strategies on experimental datasets, the following verification tasks are slated for future releases:

  • [ ] Real Experimental Dataset Benchmarking: Validate Strategy 1 and Strategy 2 on physical wind tunnel / water channel multi-camera PTV datasets with ground truth or synthetic image projections.
  • [ ] Adaptive Velocity Thresholding: Automatically infer dvxmax/dvxmin bounds per frame based on mean spatial displacement histograms prior to tracking.
  • [ ] Ensemble Consensus Voting: Implement an $N$-tracker consensus ensemble where a candidate link is accepted only if at least $M$ out of $N$ trackers agree.
  • [ ] GPU-Accelerated Bipartite Matching: Port global Hungarian cost matrix solvers to PyTorch/CuPy for $N > 50,000$ ultra-dense particle tracking.