Gerçek zamanlı hedef takibi ve ur5 işbirlikçi robot kolu ile izlenmesi

The rise of the camera usage and their availability give opportunities for developing robotics applications and computer vision applications. Especially, recent development in depth sensing (e.g., Microsoft Kinect) allows development of new methods for Human Robot Interaction (HRI) field. Moreover, Collaborative robots (co–bots) are adapted for the manufacturing industry. This thesis focuses on HRI using the capabilities of Microsoft Kinect, Universal Robot–5 (UR5) and Robot Operating System (ROS). In this particular study, the movement of a fingertip is perceived and the same movement is repeated on the robot side. Seamless cooperation, accurate trajectory and safety during the collaboration are the most important parts of the HRI. The study aims to recognize and track the fingertip accurately and to transform it as the motion of UR5. It also aims to improve the motion performance of UR5 and interaction efficiency during collaboration. One of the important parts of the study is trajectory planning. A trajectory is the path that a robot follows through as a function of time. Trajectory planning is sometimes referred to as motion planning and erroneously as path planning. Trajectory planning is distinct from path planning in that it is parametrized by time. Essentially, trajectory planning encompasses path planning in addition to planning how to move based on velocity, time and kinematics. In robotics there are inherent limitations to calculate the trajectory, especially for non–spherical wrist robots such as UR5. The computation time of a trajectory and data processing must be in milliseconds in tracking for seamless and safe HRI. Besides that, if the movements are not recognized or planned movements, safety is endangered and comfort will be lower. In the experimental part, nearest–point approach is used via Kinect sensor's depth image (RGB–D). The approach is based on the Euclidean distance which has robust properties against different environments. Moreover, Point Cloud Library (PCL) and its built–in filters are used for processing the depth data. After the depth data provided via Microsoft Kinect have been processed, the difference of the nearest points is transmitted to the robot via ROS. On the robot side, MoveIt! motion planner is used for the smooth trajectory. Once the data has been processed successfully and the motion code has been implemented without bugs, 84.18% total accuracy was achieved. After the improvements in motion planning and data processing, the total accuracy was increased to 94.14%. Lastly, the latency was reduced from 3–4 seconds to 0.14 seconds.