🔍 Real-Time Surveillance Weapon Detection

This repository presents a deep learning–powered vision system that detects and classifies weapons from live surveillance feeds or recorded videos. Unlike traditional motion-based systems, this model is trained to minimize false triggers from everyday objects (toys, kitchen items, etc.), ensuring reliable and context-aware detection. The goal is to support public safety and threat prevention by providing early alerts of dangerous objects in the scene.

🎯 Objectives

• Detect Threats: Accurately identify a wide range of weapons in live video feeds.
• Reduce False Positives: Distinguish real weapons from harmless objects for reliable alerts.
• Showcase Performance: Provide clear, standout metrics and visuals to highlight model capability.

⚙️ Installation & Setup

1. Clone the repository

   git clone https://github.com/Soohaam/Weapon-Detection-Assignment.git
   cd Weapon-Detection

2. Create & activate a virtual environment Windows (PowerShell):

   python -m venv venv
   venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

4. Download model weights The weights (best.pt) can be found in: My Drive.

   pip install -r requirements.txt
   results/weapons_train/weights/best.pt
   

📂 Project Structure

Weapon-Detection/
├── data/
│   ├── images/                # Test images
│   ├── videos/                # Test videos
│   └── weapons_dataset/       # Train/valid/test split dataset
│
├── models/                    # Base YOLO weights
│   └── yolov8n.pt
│
├── results/
│   ├── weapons_detection_results/   # Outputs (images/frames)
│   └── weapons_train/               # Training logs + weights
│       └── weights/
│           ├── best.pt
│           └── last.pt
│
├── src/
│   ├── detect.py               # Inference script (images/videos)
│   └── train.py                # Training script
│
├── requirements.txt
├── data.yaml
└── README.md

🚀 Running the Project

Run detection on an image

cd src
python detect.py ../results/weapons_train/weights/best.pt ../data/images/sample1.jpg

Run detection on a video (outputs frames in results/weapons_detection_results/)

cd src
python detect.py ../results/weapons_train/weights/best.pt ../data/videos/samplevideo.mp4

Train the model (optional, if retraining is needed) The model was trained on a custom dataset. Download it from Google Drive and place it in the data/weapons_dataset/ directory.

cd src
python train.py

📷 Dataset Sample

• 

🌟Performance Metrics & Training Results

Box F1 Curve

• 

  • Performance Highlight: Peaks at 0.56 (confidence 0.404), showcasing an optimal balance of precision and recall for threat detection.

  Box Precision Curve

• 

  • Performance Highlight: Soars to 1.0 at high confidence (0.996), ensuring near-perfect accuracy and minimizing false alarms on toys or utensils.

  Box Recall Curve

• 

  • Performance Highlight: Achieves a stellar 0.82 at low confidence, guaranteeing most genuine threats are caught.

  Box PR Curve

• 

  • Performance Highlight: Boasts an impressive mAP@0.5 = 0.543, reflecting strong overall detection across all weapon classes.

  Confusion Matrix Normalized

• 

  • Performance Highlight: Reveals a solid average accuracy, with standout detection for Knife and Missile.

📊 Model Performance Visualizations

1. mAP50(B) & mAP50-95(B) Graphs

• 
  • Description: This plot showcases the model's mean Average Precision at IoU=0.5 (mAP50) and across IoU 0.5-0.95 (mAP50-95), highlighting its robust detection capabilities over varying thresholds.

2. Precision & Recall Graphs

• 
  • Description: This plot displays the Precision and Recall curves, demonstrating the model's ability to balance accurate detections with comprehensive threat identification.

🎥 Output

📷 Image Input Results

1. Handgun Detection

Raw Image	Detected Image
Description: The original test image.	Description: A sample frame highlighting the model's accurate detection of a handgun.

2. Rifle Detection

Raw Image	Detected Image
Description: The original test image.	Description: An example frame showcasing the model's capability to detect a rifle in a real-world image setting.

🎬 Video Input Results

3. Video Output - Detection in Action

newoutput.mp4