Inspiration
This project was developed to Prevent theft, bombing attacks by terrorists, personal attacks like fighting, shooting, etc at highly sensitive areas like banks, hospitals. Video surveillance can also be used at college and school campuses to ensure safety from theft and vandalism.
What it does
The program combs through video surveillance captured by CCTV cameras and detects unusual activity like fighting. It can further contact the appropriate services like police and ambulance.
How we built it
The program is built on a long term convolutional network with multiple CNN layers and uses extremely large datasets to produce results with a 98% confidence.
Challenges we ran into
Loads of errors while creating such a large data-set with .avi files
Accomplishments that we're proud of
Coding such a large model within few hours
What we learned
Tried a different technique for tensor flow one hot encoded labels
What's next for Smart Surveillance Solutions by Simarjot
Hypertuning datasets, Real-time performance currently restricted by operating large video files on a 8GB VRAM GPU.
Built With
- keras
- ltsm
- python
Log in or sign up for Devpost to join the conversation.