Team: Liberman Jonathan, Segev Itay, Zarakhovsky Daniel.
Introduction:Our goal was creating an autonomous guarding unit which specializes in detecting and neutralizing intruders.
There has been always a need for security. The guarding unit allows any user to secure any area without the need of a human guard.
Weapon Systemhe weapon system which the guard drone uses is based on our previous project – “Wireless Arrow-Shooter”. In that project, we have cracked down an arrow shooter toy to gain a full control over it without the use of toy helicopters (Originally, this toy was designed to be added to "WLToys" toy helicopters ).
Our final result was an arrow shooting device fully controlled by us. We could choose the amount of arrows to shoot and when. The full system is light enough to be added on any drone bigger than a tablet.
For this project, we have changed the system to work on Bluetooth instead of Wi-Fi as the wireless communication protocol. We used the Bluetooth module HC-05 because It is very cheap (~ 2$) and very relient and easy to work with.
This weapon system made it possible to create other interesting projects. The idea of a drone, capable of guarding and defending any area from intruders appealed to us, and we decided to implement it.
The DroneFirst, we needed a drone to be our guard. The "AR Drone 2.0" by Parrot was chosen for its ability to stabilize itself. This drone is controlled with commands that are transmitted through the drone's Wi-Fi hotspot. We used the git repository – cvdrone by puku0x to command it through the computer (regularly it is controlled using a mobile app).
Face RecognitionFor the facial recognition, we started with OpenCV. After many experiments seeing it is very inaccurate, we switched to dlib – an open source image processing library.
With dlib our facial recognition was flawless but it took more computer resources, so a strong computer is needed.
The Main AlgorithmOur drone – the autonomous guard, as we said, needs to be fully autonomous, but still, needs to be able to protect. The drone scans the environment by spinning in place until it detects a face. When a face is detected, the drone locks on it and makes precise movements by the face location on the screen.
The cvdrone library allowed us to control the drone speed in every axle. If the target is very near the center the drone will move slowly to center the target to shoot it, if the target is very far it will move faster. When the target is centered we shoot it until it falls down.