Crime Catcher

Safety Through Vigilance!

REPO: HackRPI24-Adaptive-Threat-Security-System - Submitted via Devpost to HackRPI 2024 Urban Upgrades

What it does

Crime Catcher is a solution to keeping our urban spaces safe and secure.

It is an online surveillance system where a central server receives camera feeds from remote clients and uses AI to determine if weapons or dangerous objects are in frame. The model is GPU-accelerated for efficient processing.

If a dangerous object is detected, the backend automatically sends an email alert to a trusted contact, ensuring every second counts.

Additionally, we save and catalog videos and metadata using an SQLite database. This data can be browsed for review and analysis after the fact.

Other Features:

• Clients will automatically try to reconnect if they get disconnected from the server (they will disconnect after 5 attempts).
• We integrate time stamps into the video streams such as frame rate, source IP, source building, etc.
• Users can specify an email address in the server to receive alerts about dangerous objects and weapons.
• We used ttk bootstrap to create beautiful interfaces on both the client and server in part 1.
• For Part 2 (the vue.js/quasar and flask dashboard) we have a login system that enforces users to login before they can access the video metadata, ensuring the system is attributable
• We have motion detection in our server stream so only eventful data is saved to disk, conserving bandwidth

Technologies Used:

Challenges we ran into

• Working with Tkinter's documentation and using the CUDA framework in our server for GPU acceleration

Accomplishments that we're proud of

• Motion detection: Creating logic that pauses recording on the server if there is no motion going on in the camera frame.
• AI weapon detection: Utilizing machine learning to detect and locate weapons.
• Video feed centralization: Gathering multiple camera feeds to effectively track and stop bad actors.

What we learned

Tkinter documentation is difficult to understand, and we learned how to integrate an AI model into our Tkinter server and how to best select a pre trained model to suite our needs. We also learned more about YOLO and model training.

What's next for Crime Catcher

After our MVP, we want to iterate and add new features like remote client activation in the next sprint.

Sources:

• Text on Video with OpenCV: https://www.geeksforgeeks.org/python-opencv-write-text-on-video/
• Ascii art generated using: http://patorjk.com/software/taag/#p=display&f=Graffiti&t=Type%20Something%20
• Guns detection model from: https://www.kaggle.com/code/ahmedgaitani/guns-object-detection-code