GunShield AI: An Automated System for Gun Appearance using Human Pose
Santosh E,
Department of Computer Science and Engineering,
Maharaja Institute of Technology Mysore
Affiliated to Visvesvaraya Technological University (VTU),
Belagavi, Karnataka, India
santoshe_cse@mitmysore.in
Bhavyashree H. L,
Department of Computer Science and Engineering,
Maharaja Institute of Technology Mysore
Affiliated to Visvesvaraya Technological University (VTU),
Belagavi, Karnataka, India
bhavyashreehl_cse@mitmysore.in
Keerthi M,
Department of Computer Science and Engineering,
Maharaja Institute of Technology Mysore
Affiliated to Visvesvaraya Technological University (VTU),
Belagavi, Karnataka, India
keeerthim296@gmail.com
Niranjan J,
Department of Computer Science and Engineering,
Maharaja Institute of Technology Mysore
Affiliated to Visvesvaraya Technological University (VTU),
Belagavi, Karnataka, India
niranjanjagadeesh8@gmail.com
Krishna Yash Raj,
Department of Computer Science and Engineering,
Maharaja Institute of Technology Mysore
Affiliated to Visvesvaraya Technological University (VTU),
Belagavi, Karnataka, India
krishnayashraj15@gmail.com
Vinay Kumar J D,
Department of Computer Science and Engineering,
Maharaja Institute of Technology Mysore
Affiliated to Visvesvaraya Technological University (VTU),
Belagavi, Karnataka, India
vinayjkuppe.jd@gmail.com
Keywords: Real-Time Threat Detection,Gun Detection,Human Pose Estimation,TensorFlow,OpenCV,MediaPipe,Machine Learning,CNN (Convolutional Neural Network)
Ensuring public safety has become a major concern with the rise in firearm-related incidents. This paper presents a real-time gun and human pose threat detection system designed to identify potential threats from live or recorded video feeds. The proposed system uses TensorFlow and OpenCV for object detection and MediaPipe for analyzing human body posture to assess suspicious activities. When a possible threat is detected, the system triggers an automatic alert, captures evidence frames, and stores the event data for further analysis. The model is trained on a combination of firearm datasets and human pose images to improve accuracy in dynamic environments. Experimental results show that the system performs effectively in identifying both visible weapons and aggressive human poses under varying lighting and background conditions. This approach aims to support real-time surveillance and enhance security response through intelligent automation.
