Human Detection using YOLOv8 with Squeeze Excitation

Abstract

Human detection based on vision systems has become a crucial field in the advancement of information technology. With computer vision systems, we can detect human movements in real-time, which is a significant aspect in security and surveillance applications. One effective architecture for object detection, including humans, is YOLO (You Only Look Once). YOLO has the advantage of fast and accurate detection with a single process, enabling real-time object detection. In this research, we developed the latest YOLOv8 architecture optimized for human detection in various situations and conditions. We also utilized the squeeze-and excitation (SE) attention module to enhance human detection accuracy without significantly increasing parameters. This study aims to create a human detection system capable of achieving high accuracy and can be implemented on Jetson Nano with webcam input. The modified architecture has 4.76 million parameters, mAP 0.548, and GFLOPS 12.