Modern industrial robots are often used for manipulation of various objects. For a system to work, however, the robot needs information from its environment. An often used approach for collecting and processing of data in such systems is computer vision. In the thesis, a system was designed for recognition of the general shape, height, and position of an object, moving on a conveyor belt, with the use of an RGBD camera. The presented theory explains how 2D and depth images collected with stereo vision are formed as well as the fundamentals of the Robot Operating System (ROS). The software is developed within a ROS workspace. The software is tested in a simulated enviroment based on Gazebo and in real life using an Intel D435 RGBD camera. A comparison is made between the results from the simulated and recorded data. It is evident that the depth information is susceptible to errors from incorrect camera placement and other negative impacts from the environment. Meanwhile, the largest issue when processing 2D data is the ratio between the camera resolution and the size of the region of interest. In the end further work and system optimisation is proposed.
|
