Overview / Usage

The project is mainly focused to provide a better usage of storage and power in surveillance systems. Initializing the process of recording only upon human detection decreases the storage and power consumption of the surveillance system.

Methodology / Approach

The project is briken down into three phases, object detection, machine learning algorithms on edge devices and networking.

Object Detection: This will be a classifier that detects the presence of humans in the range of vision. On detection of a human the recording is triggered. On left idle without humans in the range of vision the recording stops and saves power and storage consumption.

Machine Learning algorithms on edge devices: The problem with machine learning models are that they tend to be bulky and hence cannot infer quick enough on an edge device. The problem is solved by using better and optimzed models and the use of Movidius Neural Compute Stick.

Networking: Chanellising different camera feeds into a centralised system needs networking. With the help of different techniques we would also want to bring down the lag in the feed and work with a lot of pings simultanously.

Technologies Used

Object Detection:

• Keras

Machine Learning algorithms on edge devices:

• Raspberry Pi
• Movidius Neural Compute Stick

Repository

https://github.com/ayulockin/optimized_surveillance

Collaborators