YOLO Security

YOLO Security is a home security application that harnesses the power of computer vision to keep the user safer and in the loop about the security of their home. It connects to the cameras around the user's house to keep an extra set of eyes on the perimeter. If the model sees a person, a car, or an animal, it will send the user a notification with an image of what it saw. This allows the homeowner to know in real-time what's happening outside their house, whether they're there or not.

Currently, the prototype is run on a web app (built with Flask, hosted on a Raspberry Pi 400), with a place where the user can upload a video of their own to see the model in action.

This product uses the YOLOv5 model, a lightweight and fast object-detection model that can run in real time. It was trained on the Pascal VOC 2007 and 2012 datasets to generate weights and optimize hyperparameters. The model is used to identify objects, classify them, and render bounding boxes on video frames as it's happening.

Challenges in the creation of this product included navigating the limitations of Google Colab (thus leading me to download CUDA and cuDNN onto my laptop), dealing with datasets, and getting my website hosted where it can be accessed externally (torch is a massive import, so Heroku couldn't do it).

As for improvements, I plan to get the web app running on a more "real" domain, with a stronger computer hosting it, and improve the weights for better performance (either via training it on the COCO dataset or through more strategic training).

Overall, I'm very proud of everything I learned how to do, and that it's a real, actually usable product! It's my first time working with computer vision and machine learning models, so I'm pleased it not only works, but it's pretty solid.

Built With

• flask
• python
• yolov5