LightMOT: a lightweight convolution neural network for real-time multi-object tracking
by Lie Guo; Pingshu Ge; Yibing Zhao; Dongxing Wang; Liang Huang
International Journal of Bio-Inspired Computation (IJBIC), Vol. 22, No. 3, 2023

Abstract: Multi-object tracking (MOT) is an important problem in computer vision with a wide range of applications. MOT networks are computationally expensive and cannot run on mobile platforms in real-time. A lightweight multi-object tracking algorithm, named LightMOT, is proposed based on fairness multi-object tracking network (FairMOT). ShuffleNet V2 is adopted as the backbone network for feature extraction. Two efficient neural network modules are designed to improve the performance of the tracking algorithm. Information fusion module (IFM) is proposed to perform feature map information fusion with a small computational cost, while information enhancement module (IEM) is used to extract feature information from the final feature map. Tests were carried out in three typical scenarios from D²-City dataset. Urban road scene tracking results show that the MOTA value increases from 51.9% to 63.2% and the ID Switch value is also reduced when adding IFM and IEM at the same time. After the model is converted to ncnn format, it can reach 30 fps on the embedded platform.

Online publication date: Thu, 14-Dec-2023

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Bio-Inspired Computation (IJBIC):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com