YOLO Security

YOLO stands for you only look once and is a process for object detection. This app uses YOLO detection to monitor a video stream and send real-time alerts.

• Advanced object detection
• Realtime E-Mail alerts

Getting Started

• If you don't already have one, make a burner E-Mail. I'd strongly recommend GMail for the purpose of uploading images to its Google photos account.
• Go to Google Security and do a find for "Less secure app access". Turn on less secure app access, this allows your dummy account to send E-Mails within the app.
• Run cp skeleton_config.py config.py and fill in the variables with the proper values
• Preferably within a dedicated virtual environment, run pip install -r requirements.txt

Using the GPU

• Run ./tf_gpu_setup.sh
• Add this to the bottom of your .bashrc

export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64"
export CUDA_HOME=/usr/local/cuda
export PATH="$PATH:/usr/local/cuda/bin"

• Instead of doing a pip install on requirements use tf_gpu_requirements.txt

For more info on setting up TF GPU TF GPU

Running the app

# This is it. I know, anti-climactic
python main.py

Todos

• Store detected object sightings in Elasticsearch
• Auto backup video into glacier
• Upload backup of discovered object capture into google pictures (cause free)
• Broadcast live stream
• Sniff for mac address & tie to image
• Set which objects to take note of
• Expand list of possible detected objects
• Allow for an offset of objects e.g. ignore 1 vehicle
• Text alerts
• Create profiles on frequently detected objects

License

It's a public repo, do whatever.