A Hybrid Image Segmentation Method for Accurate Measurement of Urban Environments

In the field of urban environment analysis research, image segmentation technology that groups important objects in landscape pixel units has been subject increased attention. However, since a dataset consisting huge amount and label pairs is required to utilize this technology, most cases, model trained with having similar characteristics used for analysis, as result, quality poor. To overcome limitation, we propose hybrid leverage strengths each predicting specific classes. particular, first introduce pre-processing operation reduce differences between collected public dataset. Subsequently, train several models pre-processed then, based on weight rule, results are fused create one map. evaluate our proposal, Google Street View images do not have any labels using cityscapes which contains foregrounds images. We quantitatively assessed its performance ground truths qualitatively evaluated GSV data through user studies. Our approach outperformed existing methods demonstrated potential accurate efficient computer vision technology.