A Review of Real-Time Deep Learning–Based Object Detection Models for Resource-Constrained Embedded Systems

Abstract

Computer vision relies heavily on object detection; from autonomous drones to industry monitoring it is everywhere. However, despite these advances when one wants to implement these cutting-edge object detection models in embedded system it proves to be difficult due to the limitations in processing power, memory and energy consumption. A detailed examination of real-time object detection models designed for resource-constrained devices is presented in this paper. We investigate widely used one stage detectors like SSD (Single Shot Multi Box) and YOLO (You Only Look Once). In addition, we also discuss model compression methods like knowledge distillation, pruning and quantization, which enable efficient deployment on embedded systems.

Conflict of Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

The  research received no  specific  grant  from  any  funding  agency  in  the  public, commercial, or not-for-profit sectors.

Data Fabrication/Falsification Statement

The author(s) declare that no data have been fabricated, falsified, or manipulated in this study.

Participant Consent

The authors confirm that Informed consent was obtained from all participants, and confidentiality was duly maintained.

Copyright and Licensing

For all articles published in the NIJEC journal, Copyright (c) of this study is with author(s).