| Title |
Recursive weighted 2-Means split algorithm for under-segmentation reduction / |
| Authors |
Magylaitė, Kornelija ; Arlauskas, Lukas ; Ryselis, Karolis |
| Full Text |
|
| Is Part of |
CEUR workshop proceedings: IVUS 2023: Information society and university studies 2023: proceedings of the 28th international conference on information society and university studies (IVUS 2023) Kaunas, Lithuania, May 12, 2023 / edited by: A. Lopata, T. Krilavičius, I. Veitaitė, A. García-Holgado.. Aachen : CEUR-WS. 2023, vol. 3575, p. 205-213.. ISSN 1613-0073 |
| Keywords [eng] |
weighted K-Means ; human body segmentation ; random forest |
| Abstract [eng] |
Human body segmentation is utilized in various applications as an intermediate step. This problem is best solved using supervised machine learning solutions, however, they require annotated data for training. Unfortunately, annotating data for segmentation is an extremely labor-intensive task. It could be improved by using semi-automatic segmentation algorithms, however, their accuracy tends to be low for complex scenes. The errors made by such algorithms can be manually corrected by a human. This process could be more efficient when automatic correction tools are added to the data processing pipeline. This research aims to improve the final semi-automatic segmentation accuracy by improving an existing random forest classifier for correcting point cloud segmentation based on metrics of recursive 2-Means split. We replace the K-Means clustering with a weighted K-Means clustering and optimize the weights. Experiments have revealed that the segmentation accuracy of 62.4% is improved to 66.1% with a weight ratio of 0.4. Since higher accuracy leads to less manual labor, this is a sought-after improvement that reduces the time to prepare datasets for human body segmentation. |
| Published |
Aachen : CEUR-WS |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2023 |
| CC license |
|
