DEVELOPMENT OF HARDWARE-IN-THE-LOOP SIMULATION FOR VISUAL TARGET TRACKING OF OCTOROTOR UAV

• Main Author: TRIADHITAMA (NIM : 13206064) ; pembimbing :Prof. Dr. Ir. Bambang Riyanto Trilaksono ; Widyaward, RYAN
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/17121
• Institution: Institut Teknologi Bandung

The purpose of digital image processing is to retrieve useful information contained in the captured image from instruments and use that information for decision making to complete a tasks as if the system is able to see, which is typically called computer vision system. One is to recognize and determine the position of an object, on a captured image from camera. This is done by using an appropriate algorithm, hence the features of the object can be extracted and compared with the existing knowledge, therefore the object can be identified, as how humans recognize objects. This algorithm produces robust features those invariant to scale, rotation, illumination, and perspective. By combining the computer vision system with an <br /> <br /> <br /> <br /> <br /> unmanned system, octorotor UAV for example, extracted information can be used as reference data for navigation. Therefore, the UAV can mimic how humans in performing target tracking of an object which is considered as target. However, an appropriate testing of designed hardware must be done under few condition to analyze the performance and then make conclusions whether it is feasible to be applied into the plant. &#65532;The designed computer vision system correlates the extracted features from reference image in memory and captured image from a camera using Speed-Up <br /> &#65532;Robust Features (SURF) algorithm. &#65532;SURF detects feature <br /> &#65532;keypoints by searching blob-like structure, based on the determinant of Hessian matrix. By using integral image representation, computation can be done faster. Detected feature is subsequently described by descriptor vectors which describes iv &#65532;gradation from black to white (low to high intensity value) at points around the &#65532;feature. These features are then matched using Fast Library for Approximate Nearest Neighbor (FLANN) algorithm. With this algorithm, the the data are structured in a multiple randomized kd-tree to be processed in parallel. Approximation of Nearest Neighbor search is done by limiting the amount of leafs that are searched. The center of mass of correlated features is then calculated to determine the position the target. Target must always be in the center of the image plane. Target position error gives command to oktorotor movement, forward or backward <br /> <br /> <br /> <br /> <br /> within a certain speed, and turn left or right in a certain angular speed. <br /> <br /> <br /> <br /> <br /> Hardware-in-the-loop simulation (HILS) is done by using existing navigation and flight controller structure on octorotor, and model of octorotor dynamics that has been obtained. Data from computer vision system, which is target position in image plane, is sent via UDP protocol. The dynamics of the UAV and environment are subsequently visualized and displayed on the screen to be captured by the camera. UAV simulation is done using Matlab Simulink and visualization of UAV dynamics and environment is done using FlightGear. Target tracking simulation shows satisfactory result of performance of visual target tracking of moving car at speed around 8 knots or 15 km/h in light maneuver. Navigation and flight control system is able and adequate to perform target tracking. Hence, computer vision system can be applied on the UAV oktorotor.