Scholarlink

In today’s world, where security concerns, are on the rise it has become more important than ever to detect weapons in environments. Our focus is on identifying individuals who are carrying handguns in forms of media such as images, videos, and CCTV footage. In this work, we offers a solution by utilizing learning techniques specifically using the YOLOv8 Nano model. To train and validate the model we have created a dataset consisting of 16,000 images featuring handguns and people holding them. The principal purpose of this study is to assess the ability of existing model to quickly and precisely detect threats in surveillance settings. The proposed work paves the way to enhance the capabilities of security systems allowing law enforcement agencies to promptly identify weapons and minimize response times while mitigating risks. More specifically, we proposed a dataset that was collected from several internet sources and the dataset posses most of the real time challenges in weapon detection such as blurrness of small weapons, partially occluded weapons, weapons having shape similarity with safe objects. The dataset collection is an ongoing process, in this work we have taken a part of the dataset and validate the dataset using YOLOV8 Nano Deep learning model.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning Based Small Object (Weapon) Detection in Complex Scene Using YOLO-V8

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Navuluri Hemanth Srivathsav,
  • Vegesh Sai Boppana,
  • Rajib Debnath,
  • Kakali Das

摘要

In today’s world, where security concerns, are on the rise it has become more important than ever to detect weapons in environments. Our focus is on identifying individuals who are carrying handguns in forms of media such as images, videos, and CCTV footage. In this work, we offers a solution by utilizing learning techniques specifically using the YOLOv8 Nano model. To train and validate the model we have created a dataset consisting of 16,000 images featuring handguns and people holding them. The principal purpose of this study is to assess the ability of existing model to quickly and precisely detect threats in surveillance settings. The proposed work paves the way to enhance the capabilities of security systems allowing law enforcement agencies to promptly identify weapons and minimize response times while mitigating risks. More specifically, we proposed a dataset that was collected from several internet sources and the dataset posses most of the real time challenges in weapon detection such as blurrness of small weapons, partially occluded weapons, weapons having shape similarity with safe objects. The dataset collection is an ongoing process, in this work we have taken a part of the dataset and validate the dataset using YOLOV8 Nano Deep learning model.