Research on YOLOv4 model inference acceleration of real time object detection for smart construction site

Jianchun Wang; Minjian Long; Yunfu Zhou; Congcong Guan

doi:10.1117/12.2653817

18 November 2022 Research on YOLOv4 model inference acceleration of real time object detection for smart construction site

Jianchun Wang, Minjian Long, Yunfu Zhou, Congcong Guan

Author Affiliations +

Proceedings Volume 12473, Second International Conference on Optics and Communication Technology (ICOCT 2022); 124731B (2022) https://doi.org/10.1117/12.2653817
Event: Second International Conference on Optics and Communication Technology (ICOCT 2022), 2022, Hefei, China

Abstract

In smart construction site, multi objects in real time monitoring streams are often needed to be detected at the same time. If YOLOv4 models are not accelerated, higher inference delay will be occurred, so that the purpose of real time detection can’t be achieved. The features of YOLOv4 model are firstly introduced in this paper, and then we discuss how to use YOLOv4-tiny-3l and TensorRT to accelerate the inference process of YOLOv4 model in detail. The experiments show that YOLOv4-tiny-3l models can be used to detection objects in multi real time streams smoothly, but the accuracy is pretty poor, so that the models can’t be used in practices. When adopting TensorRT toolkit to quantize YOLOv4 models with FP16 precision, the accelerated models can be used to detect objects in multi real time streams smoothly with a small loss of accuracy.

Citation Download Citation

Jianchun Wang, Minjian Long, Yunfu Zhou, and Congcong Guan "Research on YOLOv4 model inference acceleration of real time object detection for smart construction site", Proc. SPIE 12473, Second International Conference on Optics and Communication Technology (ICOCT 2022), 124731B (18 November 2022); https://doi.org/10.1117/12.2653817

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