Autonomous UAV with vision based on-board decision making for remote sensing and precision agriculture

Abstract

In recent years, a phenomenal increase in the development of Unmanned Aerial Vehicles (UAVs) has been observed in a broad range of applications in various fields of study. Precision agriculture has emerged as a major field of interest, integrating unmanned monitoring of crop health into general agricultural practices for researchers are utilizing UAV to collect data for post-analysis. This paper describes a modular and generic system that is able to control the UAV using computer vision. A configuration approach similar to the Observation, Orientation, Decision and Action (OODA) loop has been implemented to allow the system to perform on-board decision making. The detection of an object of interest is performed by computer vision functionality. This allows the UAV to change its planned path accordingly and approach the target in order to perform a close inspection, or conduct a manoeuvres such as the application of herbicide or collection of higher resolution agricultural images. The results show the ability of the developed system to dynamically change its current goal and implement an inspection manoeuvre to perform necessary actions after detecting the target. The vision based navigation system and on-board decision making were demonstrated in three types of tests: ArUco Marker detection, colour detection and weed detection. The results are measured based on the sensitivity and the selectivity of the algorithm. The sensitivity is the ability of the algorithm to identify and detect the true positive target while the selectivity is the capability of the algorithm to filter out the false negatives for detection targets. Results indicate that the system is capable of detecting ArUco Markers with 99% sensitivity and 100% selectivity at 5 m above the ground level. The system is also capable of detecting a red target with 96% sensitivity and 99% selectivity at the same height during a test height at 5 metres. This system has potential applicability in the field of precision agriculture such as, crop health monitoring, pest plant detection which causes detrimental financial damage to crop yields if not noticed at an early stage.