📅 Aug 2025

https://github.com/Pushkkaarr/Weapon-Detection-Using-Fast-R-CNN

WHAT IT DOES

Real-time weapon detection system for surveillance applications. Processes images and video feeds to automatically identify and localize firearms with bounding box annotations and evidence frame extraction.

TECH STACK

Deep Learning: Faster R-CNN (Region-based CNN), PyTorch/TensorFlow [specify]

Computer Vision: OpenCV, PIL/Pillow

Dataset: 14,000+ JSON-annotated images (COCO format)

Training: Transfer learning from pre-trained COCO weights

Deployment: Python Flask/FastAPI for API serving [specify if applicable]

Inference: Real-time video processing pipeline

KEY FEATURES

Object Detection Model – Built using Faster R-CNN with transfer learning from COCO pre-trained weights, trained on 14,000+ annotated images to detect firearms with precise bounding boxes and confidence scores, achieving 82% [email protected].

Real-Time Video Inference Pipeline – Processes live or recorded surveillance footage frame-by-frame with adjustable confidence thresholds, renders bounding boxes on detected weapons, and extracts critical evidence frames for security review.

Image Processing & Batch Inference – Supports batch image detection with JSON-formatted outputs containing bounding box coordinates and confidence values, along with preprocessing steps like normalization, resizing, and augmentation.

METRICS

APPLICATIONS

Designed for surveillance and security systems in public spaces, schools, airports, and high-security environments, enabling automated firearm detection, real-time threat alerts, and evidence preservation for post-incident analysis.