Application of Machine Vision in Unmanned Aerial Systems for Autonomous Target Tracking

Abstract

This paper presents experimental results for an application of Machine Vision (MV) techniques for addressing the problem of target detection and tracking related to unmanned systems. The main objective was to design a prototype surveillance environment to emulate real-life conditions onboard an aerial vehicle. The environment for this experiment consisted of a target - small electric train - located at “ground” level and a motion-based MV camera apparatus located above the model train assembly. For detecting and tracking the movements of the target, MV algorithms were implemented in a Matlab/Simulink ® based simulation enviroment. Specifically implementing image pre-processing techniques and circle detection algorithms for identifying specific features of the moving train. A comparative study was performed between the different MV algorithms utilizing several trials with the target moving along different directions. A performance assessment was completed based on the detection error; an analysis of the computational workload for each algorithm is also presented. Results indicate a solution path towards the development of a vision-based control strategy for allowing unmanned systems to participate in complex missions involving aspects of autonomous target tracking.