Object Detection and Pose Estimation in 3D Images

U ovom radu istražene su i implementirane dvije metode prepoznavanja objekta od interesa u svrhu određivanja putanje robotskog manipulatora po površini objekta. Implementirane su LINE-MOD i tabletop metode prepoznavanja objekta. Objekt od interesa je zadan u obliku mreže trokuta. Metode za prepoznavanje objekata su implementirane u ROS okruženje te koriste informacije dobivene od RGB-D kamere za prepoznavanje. Kako bi se mogla planirati putanja robota po površini objekta, potrebno je znati gdje se nalazi objekt u odnosu na koordinatni sustav robota. U tu svrhu je u programskom rješenju implementirana hand-eye kalibracija. Na kraju je provedeno testiranje sustava koji se sastoji od industrijskog robota ABB IRB 2400L i Orbbec Astra S kamere kako bi se odredila točnost cijelog sustava.In this thesis, two methods for object recognition are examined and implemented for the purpose of determining the path of a robotic manipulator end effector along the object surface. LINE-MOD and tabletop methods are implemented. The developed program requires the object of interest to be provided in the form of a triangular mesh. Methods for object recognition are implemented in ROS framework and they use information obtained from an RGB-D camera. In order to be able to plan the path of a robot end effector along the surface of an object, it is necessary to know where the object is located relative to the coordinate system of the robot. For that purpose, hand-eye calibration was implemented in the software solution. Finally, testing of the ABB IRB 2400L robot and Orbbec Astra S camera system was performed to determine the accuracy of the system