Aegis

Slide deck

By:

Aiden C

Kevin X

​

Table of contents

01

05

04

02

03

Problem

Frontend

How it works / Tech Stack

Mission

Backend

Problem

01

School-related violence with injuries by guns and knives are a critical concern.

​

In 2021, more than 250 incidents occurred nationwide on school grounds related to this type of weapon.

​

Knife-related assaults still remain at critical status, with about 4,000 adolescents treated annually from this type of injury.

​

In fact, in our very own HS, we had 2 bomb threats just this year.

– Mission

02

AT AEGIS

• Advanced AI and computer vision technologies to make schools safer.
• Develop cutting-edge early detection systems for firearms and knives to identify potential threats before they escalate.

​

We continue to advance our technologies so that every school, big or small and with any kind of resources, can access leading safety solutions, making it possible to provide a secure learning environment for students and staff.

​

​

Backend

03

YOLO v8 detects weapons and Depth Anything V2 estimates depth in video frames to identify threats.

Python Rest API processes videos and returns annotated results and detection data.

Cloud GPUs accelerate video analysis, combining detection and depth estimation for rapid threat identification.

Real-Time Object Detection and Depth Estimation

Python REST API Integration

Cloud-Based Scalable Processing

Frontend

04

In the development of this product, I used four tools:

• Shadcn UI Library - a component library
• Tailwind CSS - a CSS framework
• Framer Motion - a motion library for React
• Figma - a collaborative web application for interface design

​

These tools primarily helped me develop quicker and add interactive features to the website. In addition to these tools, I also used seesaw, a web design inspiration website.

Tailwind CSS

Tailwind CSS is a CSS framework that allows you to quickly apply CSS styles through classes.

Framer Motion

Framer Motion is a React motion library that allows you to add animations and interactions to React code.

How It Works (brief)

05

​

• Uses YOLOv8 for weapon detection and Depth Anything V2 for depth estimation

• Processes video frames on cloud GPUs with CUDA acceleration

• Outputs annotated video, depth maps, and JSON statistics

• Deployed in Docker on Paperspace with optimized async processing

​

​

YOLOv8

We used YOLOv8 on our modified custom dataset of over 20k images on Google Colab. Experimenting with the accuracy, we managed to achieve an accuracy of 0.93 at 48 epochs before the accuracy appeared to stagnate.

A little bit about YOLO: YOLOv8 is designed with a strongly modified CSPDarknet backbone, PANet neck for aggregating features at different scales, and a decoupled head. It takes the anchor-free detection approach and predicts object centers and box sizes. The model uses the Feature Pyramid Network (FPN) for multi-scale detections and applies Non-Maximum Suppression (NMS) in post-processing to clean up duplicates from detections. High-performance inference is optimized by YOLOv8 and is trained using a combination of classification, localization, objectness, and IoU losses.

​

​

YOLO v8 Model

Object Detection & Depth Estimation:

Depth Anything V2 Model

React JS

Frontend Technologies

Tailwind CSS & Shadcn UI Library Framer Motion, Figma.

Python REST API Firebase DB

Backend Infrastructure

Cloud GPU Hosting (Paperspace)

DigitalOcean

Website Hosting

Docker

Deployment & Management:

05

How It Works / Tech Stack

Conclusion

Despite being red, Mars is actually a cold place. It’s full of iron oxide dust

Venus has a very beautiful name and is the second planet from the Sun

Mercury is the closest planet to the Sun and the smallest one in the Solar System

Neptune is a ig planet. It is the fourth-largest planet by diameter in the Solar System

Mars

Venus

Mercury

Neptune