Landscape Design and Green Space Perception Enhancement Based on Panoramic Imaging and Segnet

Abstract

Green view rate is an intuitive evaluation criterion used for green space perception. Traditional research on green view rate is mostly calculated based on flat images, which cannot fully reflect the subjective human perception of green volume in three-dimensional space. Based on the panoramic image, we propose the concept of panoramic green perception rate, obtain spherical panoramic photos by panoramic camera, convert isometric cylindrical projection into isoprojective cylindrical projection, and use the convolutional neural network model based on semantic segmentation to automatically identify the area of vegetation to achieve automatic recognition and measurement of panoramic green perception rate. The results were compared with manual discrimination and showed that the average intersection ratio (mIoU) of greenery recognition using Dilated ResNet-105 was 62.53%, and the average difference with manual recognition was 9.17%.