|
|
3D point cloud segmentation for industrial bin-picking
Šooš, Marek ; Škrabánek, Pavel (oponent) ; Shehadeh, Mhd Ali (vedoucí práce)
This thesis deals with 3D point cloud segmentation for industrial bin-picking, a key challenge in the field of industrial robotics. The aim of the thesis is to develop and deploy a highly effective algorithm for segmenting and registering 3D point clouds, thereby improving the accuracy, speed, and efficiency of bin-picking operations. The contribution of the thesis is the presentation of the researcher's solution to create artificially generated data needed for training. The thesis results in a symbiosis of advantages of a fast-segmentation algorithm based on machine learning, and a high quality, robust but slow algorithm based on geometric principles. Functionality, reliability and quality of the developed algorithm were also experimentally verified on different types of objects and different datasets. Results of the work show that the proposed algorithm yields a fast, reliable, and comprehensive solution to the bin-picking problem. Customized data generation reduces the time and cost of applying such a system. In conjunction with a comprehensive problem solving system we are able to accurately and easily generate applications for diverse and specialized bin-picking tasks. Achieved results contribute to the development of point cloud segmentation methods and their applications in various industrial and scientific fields. By putting the proposed system into practice we significantly contribute to performance and reliability of the proposed automatic line.
|
host ::
RSS.