VISION-BASED OBJECT TRACKING FOR AN UAV ROTORCRAFT USING ROBOT OPERATING SYSTEM (ROS)

Abstract

In agricultural activities, it is difficult to monitor and protect the crops in a large open space. This leads to the need of a high number of workforces to perform the task. This problem may be alleviated with the application drones with machine vision feature to track any intruders, including thieves and any animals that may destroy the crops such as wild boars. This paper presents the implementation of Robot Operating System (ROS) for object tracking in Unmanned Aerial Vehicles (UAV) rotorcraft. Six object trackers in the OpenCV library including BOOSTING Tracker, Multiple Instant Learning (MIL) Tracker, Kernelized Correlation Filter (KCF) Tracker, Tracking, Learning and Detection (TLD) Tracker, Medianflow Tracker and Minimum Output Sum of Squared Error (MOSSE) Tracker, are tested and compared on the laptop and Raspberry Pi to determine the best object tracker. The efficiency of the object trackers is determined by its frame per second (FPS) and its ability to keep on tracking the object continuously. The results show that MOSSE provides the best performance and it has been programmed on the Raspberry Pi and then attached to the drone to perform the object tracking task. The outcome verifies that the UAV is able to perform the object tracking task as desired using Robot Operating System (ROS).