SMaRT: Stick via Motion and Recognition Tracker

This paper presents SMaRT (Stick via Motion and Recognition Tracker), a novel multi-object tracking (MOT) approach that integrates motion estimation and re-identification within a unified, efficient framework. Inspired by leading MOT methods like CenterTrack and FairMOT, SMaRT enhances tracking robu...

Full description

Saved in:
Bibliographic Details
Main Authors: Fatih Emre Simsek, Cevahir Cigla, Koray Kayabol
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11003088/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper presents SMaRT (Stick via Motion and Recognition Tracker), a novel multi-object tracking (MOT) approach that integrates motion estimation and re-identification within a unified, efficient framework. Inspired by leading MOT methods like CenterTrack and FairMOT, SMaRT enhances tracking robustness by fusing re-identification features from an advanced teacher-student model. This integration enables the simultaneous regression of object locations and extraction of re-identification vectors within a single neural network. Evaluations on the DIVOTrack, MOT17 and SOMPT22 datasets demonstrate significant improvements over previous state-of-the-art methods in terms of Higher Order Tracking Accuracy (HOTA), Multi-Object Tracking Accuracy (MOTA), and Association Accuracy (AssA). Additionally, SMaRT’s efficiency and accuracy are validated through comprehensive synthetic video experiments, highlighting its adaptability to varied motion patterns and occlusions. The proposed approach offers a robust, accurate, and efficient solution for real-world applications such as surveillance, autonomous driving, and robotics. The tracker is available at: github.com/sompt22/SMaRT.
ISSN:2169-3536