Scholarlink

Crime and delinquency are currently among the greatest challenges, highlighting the need to integrate advanced technologies into urban security systems. In the present study, an efficient and accurate system was developed to detect weapons (guns and knives) in real-time in surveillance videos using Computer Vision techniques. The applied methodology was based on 5 phases: obtaining the dataset; Preprocessing (Resizing and Grayscale); Feature extractors (BOVW, HOG and SIFT); Models Machine and Deep Learning models such as (KNN, SVM, DT, RF, AdaBoost, VGG16 and CNN). The proposed system demonstrated high accuracy in detecting weapons with the VGG-16 model: Accuracy of 0.9802, Precision of 0.9802, and Recall of 0.9802, reliably differentiating between threatening and non-threatening objects. Deploying this technology in high-risk areas, such as airports and public events, could significantly improve security measures by providing fast and accurate alerts. In future, we plan to focus on optimizing these algorithms for low-cost devices and extending detection to a wider range of threats.

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

Robust Model for Classification and Detection of Knives and Guns Using Computer Vision

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jeferson Sandoval,
  • Christopher Echenique,
  • Diego Rosales,
  • Victor Ormeño,
  • Junior Fabián,
  • Wilfredo Ticona

摘要

Crime and delinquency are currently among the greatest challenges, highlighting the need to integrate advanced technologies into urban security systems. In the present study, an efficient and accurate system was developed to detect weapons (guns and knives) in real-time in surveillance videos using Computer Vision techniques. The applied methodology was based on 5 phases: obtaining the dataset; Preprocessing (Resizing and Grayscale); Feature extractors (BOVW, HOG and SIFT); Models Machine and Deep Learning models such as (KNN, SVM, DT, RF, AdaBoost, VGG16 and CNN). The proposed system demonstrated high accuracy in detecting weapons with the VGG-16 model: Accuracy of 0.9802, Precision of 0.9802, and Recall of 0.9802, reliably differentiating between threatening and non-threatening objects. Deploying this technology in high-risk areas, such as airports and public events, could significantly improve security measures by providing fast and accurate alerts. In future, we plan to focus on optimizing these algorithms for low-cost devices and extending detection to a wider range of threats.