Abstract:
Our project deals with the real time collision avoidance between two UAV’s
in real time. The solution for this problem is divided into five parts. The first and
foremost part is controlling the path of drone using either ROS (Robotic
Operating System) or Python. The next step is detection of nearby UAV. For this
purpose we have used onboard sensor (camera) and deep learning methodologies.
The third step is state estimation in which we find the position and distance of
detected drones to find out where the drone is present. We have used Sonar
sensors to measure the distance of nearby UAV. The fourth step is to estimate the
chance of collision and finally if the collision is detected we need some collision
avoidance mechanism.
Collision between two objects when either one of the object is static is a
relatively simple problem and has already been dealt with, but our problem
involves collision between two moving objects which is a far complicated
problem and involves various factors. In this report, we’ll mention what those
factors are and how we are able to deal with some of them.
We have used multiple approaches in each step. For example state estimation
is also done by using sensors and computer vision based approach. We have listed
these multiple approaches and also the challenges we faced in each approach and
why we have selected one particular approach.
We also present a data augmentation technique we applied for generating the
dataset, which is then used for drone detection and also made this dataset
available to use for public.The technique is quite simple and generic and can be
used for generating any type of such simple datasets. Results and conclusions also
mention the challenges we faced in the implementation of this project and how
the stability of UAV is a major factor in making the UAV autonomous and
deploying collision avoidance on actual UAV.
