Gun Detection Dataset

枪支暴力在世界上是一个严重的问题，尤其是在美国。已经研究了深度学习方法来检测监视摄像机或智能IP摄像机中的枪支，并向安全人员发送实时警报。枪支检测算法开发的一个问题是缺乏大型公共数据集。在这项工作中，我们首先发布一个数据集，其中包含用于枪支检测的51k带注释的枪支图像，以及我们从几个不同来源收集的用于枪支分类的其他51k带裁剪的枪支芯片图像。据我们所知，这是研究枪支检测的最大数据集。我们提出了一种枪支检测系统，该系统使用智能IP摄像机作为嵌入式边缘设备，并使用云服务器作为设备，数据，警报的管理器，并进一步降低误报率。我们研究寻找嵌入式设备中枪支检测以及边缘设备和云服务器上枪支分类的解决方案。这种边缘/云框架使得在现实世界中部署枪支检测成为可能

Gun violence is a serious global problem, particularly in the United States. Deep learning methods have been explored for detecting firearms in surveillance cameras or smart IP cameras, with real-time alerts sent to security personnel. A key challenge in developing firearm detection algorithms is the lack of large-scale public datasets. In this work, we first release a dataset consisting of 51k annotated firearm images for firearm detection, plus an additional 51k cropped firearm chip images collected from multiple distinct sources for firearm classification. To the best of our knowledge, this is the largest dataset dedicated to firearm detection research. We propose a firearm detection system that employs smart IP cameras as embedded edge devices, and a cloud server as the manager for devices, data and alerts, which further lowers the false positive rate. We investigate solutions for firearm detection on embedded devices, as well as firearm classification on both edge devices and cloud servers. This edge/cloud framework enables the practical deployment of firearm detection systems in real-world scenarios.

搜集汇总

背景与挑战

背景概述

Gun Detection Dataset是一个专注于枪支检测的大型公共数据集，包含总计102k的枪支图像（51k带注释图像和51k裁剪图像），由深圳神目信息技术有限公司于2020年发布。该数据集支持深度学习算法开发，旨在通过智能IP摄像机和云服务器实现实时枪支检测和警报系统。