Publications / 2016 Proceedings of the 33rd ISARC, Auburn, USA
This paper presents the applicability of a nonparametric scene parsing model for recognizing all objects in an image. The data-driven model labels all the pixels of query images as their object classes using the labels in similar images of the existing dataset. The model has a flexible number of parameters depending on the size of the training data, it is possible to add or remove new data without re-training the whole model. The capability of the nonparametric model would improve the monitoring performance on the construction site with updating the size of the image dataset over time.