Overview:

Rapid expansion of metro cities has given rise to security concerns and has resulted in increased need of intelligent CCTV cameras to ensure the safety of people. The activity recognition of videos is a complex process. We have came up with an approach to tackle the security issue with deep neural networks.

Problem Statement:

We need to address the manual monitoring of CCTV cameras around a metro city and replace it with more robust and automated solution such as some intelligent anomaly detection agent.

Objective:

The main and prime objective for this project is to help locate the anomalous event. To achieve that , we need to train the 3D neural network model. This will help authorities to cop up against the crimes and incidents happening around the city.

Challenges:

• There is a negligible borderline between normal and abnormal activities. The same normal event can be represented as abnormal activity under certain conditions such as crowded scene.

• The abnormal actions vary form place to place and region to region , so the dataset needs to be diverse.

• The point to view for a cameras is a contributing factor for correct anomaly detection.

• There is always a possibility of noise in data , which ultimately leads to wrong predictions.  

Tools and Technologies:

Tools and technologies that will be used for this project, but might not limited to, include: 

Software Technologies and Frameworks:

For deep neural network:

• NVIDIA Deepstream

• NVIDIA CUDA

• Tensorflow

• PyTorch

• OpenCV

• Google Colab

• Linux

• Virtual environments

• 3D and 2D CNN or RNN

For Web and Mobile Interfaces (UI):

• Flask

• Node.js

• Angular.js

• Bootstrap

• NoSQL/MySQL

• Google cloud

• Heroku/AWS EC2 / AWS ELB

Demo Videos: