-
-
Sample1
-
Sample2
-
Sample3
-
Sample4
Inspiration
Our inspiration stemmed from the rising violence and wars across the world, particularly in conflict zones like India/Pakistan and Russia/Ukraine. We were driven by the need to protect innocent civilians and assist security forces in identifying threats more effectively. SafeScope was started out of our desire to use technology for peacekeeping — helping distinguish between civilians, militants, and armed personnel through intelligent image analysis.
What it does
The main usage of the application is the automated identification of illegal weapon possession in conflict areas. SafeScope is an AI-powered detection system that analyzes images to detect firearms and flag individuals who are not in uniform but appear armed. When unauthorized weapon possession is detected, the system generates a comprehensive alert package for transmission to security monitoring platforms. Each alert contains a captured image with bounding box annotations highlighting the detected weapon, confidence scores for both weapon and uniform detections (when applicable), location data derived from the camera's registered position, timestamp of the detection event, and IoU calculation (when both weapon and uniform are detected).
How we built it
We developed SafeScope using Python and Streamlit for a fast, intuitive web interface. The system leverages two custom-trained YOLOv8 models: one for detecting guns and another for classifying people based on attire (uniformed vs. non-uniformed). OpenCV handles image processing, and we implemented IoU-based logic (Area of Intersection)to detect overlaps between weapons and people.
Challenges we ran into
One major challenge was fine-tuning the object detection models to minimize false positives — especially distinguishing between similar objects and attire. Environmental variability raises serious issues since performance might suffer greatly in extreme weather events including heavy rain, dense fog, low light scenarios, or dust storms frequent in desert border areas. These disorders might compromise detection dependability by changing the visual qualities of both weapons and uniforms and so reducing visibility. Another important difficulty is handling congestion, especially when weapons are purposefully hidden by carriers trying to evade detection or partially covered behind objects.
Accomplishments that we are proud of
This work presents a novel method based on cascading YOLOv8 architecture for illegal weapon detection in conflict areas. Our system achieves both computational economy and contextual awareness by separating the detection process into specialized models for weapons and army uniforms. By means of IoU computations, the spatial relationship analysis offers a strong means of differentiating between authorized and illegal weapon possession
What we learned
Through this project, we deepened our understanding of YOLO-based object detection, image annotation, and web-based deployment using Streamlit. We also learned the importance of UI design in making complex systems accessible and user-friendly. Finally, we learned how to apply AI ethically in a real-world context, balancing safety, accuracy, and usability.
What's next for SafeScope
We plan to expand SafeScope to support real-time video analysis, making it useful for surveillance cameras and drones. We also aim to integrate geolocation tagging and incident reporting features for deployment in security operations. Future versions could include multilingual support, mobile compatibility, and integration with facial recognition and behavioral analysis to enhance situational awareness in conflict zones.
