MotionFormer: multi-object tracking with motion information

Cheng Fu; Lihui Cen

doi:10.1117/12.3011539

1 December 2023 MotionFormer: multi-object tracking with motion information

Cheng Fu, Lihui Cen

Proceedings Volume 12940, Third International Conference on Control and Intelligent Robotics (ICCIR 2023); 129402R (2023) https://doi.org/10.1117/12.3011539
Event: Third International Conference on Control and Intelligent Robotics (ICCIR 2023), 2023, Sipsongpanna, China

Abstract

In this work, we propose MotionFormer, a multi-object tracking method incorporating motion information. MotionFormer is a Transformer-based architecture, which is an attention-based query-key mechanism. A dynamic Kalman filter is proposed to predict the position when the object is occluded and the appearance characteristics are destroyed. TIoU (Tracking Intersection over Union) uses motion to improve the IoU distance measure, making it possible to focus on effective location information around objects. On MOT17 benchmark, MotionFormer achieves 74.9% MOTA.

(2023) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Cheng Fu and Lihui Cen "MotionFormer: multi-object tracking with motion information", Proc. SPIE 12940, Third International Conference on Control and Intelligent Robotics (ICCIR 2023), 129402R (1 December 2023); https://doi.org/10.1117/12.3011539

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available