← Back ICRA 2026

HybridTrack: A Hybrid Approach for Robust Multi-Object Tracking

Leandro Di Bella, Yangxintong Lyu, Bruno Cornelis, Adrian Munteanu

PDF

AI summary

Key figure (auto-extracted from paper)

HybridTrack achieves state-of-the-art accuracy and real-time speed for 3D vehicle tracking by learning Kalman filter parameters directly from data, eliminating manual tuning.

Multi-Object Tracking Kalman Filtering Autonomous Driving Deep Learning Real-Time Tracking 3D Tracking

Problem

Traditional model-based multi-object tracking relies on predefined motion models and manual noise parameter tuning, limiting adaptability to dynamic traffic scenarios, while fully deep learning-based methods often sacrifice computational efficiency and interpretability.

Approach

HybridTrack integrates a learnable Kalman filter into a tracking-by-detection framework, using lightweight neural networks to dynamically predict transition residuals and Kalman gains directly from data, removing the need for scene-specific manual design.

Key results

82.72% HOTA accuracy on the KITTI dataset
112 FPS processing speed enabling real-time deployment
Elimination of manual motion and stochastic parameter tuning
Dynamic scaling mechanism for stable initialization and missed detections

Why it matters

It delivers a robust, adaptable, and computationally efficient tracking solution critical for real-world autonomous driving and ADAS systems.

Abstract

The evolution of Advanced Driver Assistance Sys- tems (ADAS) has increased the need for robust and generaliz- able algorithms for multi-object tracking. Traditional statistical model-based tracking methods rely on predefined motion models and assumptions about system noise distributions. Although computationally efficient, they often lack adaptability to varying traffic scenarios and require extensive manual design and pa- rameter tuning. To address these issues, we propose a novel 3D multi-object tracking approach for vehicles, HybridTrack, which integrates a data-driven Kalman Filter (KF) within a tracking- by-detection paradigm. In particular, it learns the transition residual and Kalman gain directly from data, which eliminates the need for manual motion and stochastic parameter modeling. Validated on the real-world KITTI dataset, HybridTrack achieves 82.72% HOTA accuracy, significantly outperforming state-of- the-art methods. We also evaluate our method under different configurations, achieving the fastest processing speed of 112 FPS. Consequently, HybridTrack eliminates the dependency on scene- specific designs while improving performance and maintaining real-time efficiency.1

Index terms

Intelligent Transportation Systems Visual Tracking Motion Control