Capture what you are not able to

Motivation

Recently, school safety and security has become a great issue, many incidents have taken place in or near school or campus. By improving the current surveillance camera system, we can prevent and eliminate these unpleasant incidents.

Thus, we decide to implement a system that can help us detect human from surveillance camera, and make sure these possible suspects is not too close to restricted area. Once they invade the restricted area, our system can quickly notice their trespassing, and notify security to take action.

Implementation

1. Perform object detection from images input

To perform the object detection task, we choose darknet as our deep learning framework, and build our detector based on their model yolo.

2. Depth estimation

In our final system, we choose the Bounding box approach due to some limitations of TX1 developement board. We will discuss the detail in the following subsection.

3. Alert when suspects are getting too close to restriced area.

the monitor screen will gradually turn red when object starts to get close to restricted area. This will attract security/guard’s attention, let them be able to monitor this possible intruder eariler.

(Disclaimer : The picture below is taken by sialicence.co.uk. We use it only to demo our system for academic purpose and we do not own the copyright of it.)