Computers and Electronics in Agriculture
Point-based rendering optimization with textured meshes for fast LiDAR visualization
Computers & Geosciences
| Hi-index | 0.00 |
In recent years, an increase has occurred in the use of new sensors such as digital measurement cameras or LiDAR (Light Detection And Ranging) devices in remote sensing studies. This situation has encouraged the development of new possibilities for data use. Currently, it is common to perform combined flights where digital cameras take images that record spectral information while LiDAR sensors produce point clouds with positional, spectral and echo information. The goal of this study was to assess the possibility of combining LiDAR intensity with the spectral information provided by digital cameras to increase crop classification accuracy. Due to the geometric characteristics of the LiDAR data collection process, LiDAR intensity was normalized before combining it with the spectral information from the camera. Two different geometric methods were used for this purpose. Both methods were based on the relative position of each point and the position of the sensor at the time of data recording. To analyze the effects of normalization, land-use samples with different radiometric behaviors were taken. The samples were used to assess the degree of variation in the intensity values of a same geographic area acquired in different flight strips by comparing the coefficient of variation of raw intensity data and normalized intensity data. Supervised classifications with the maximum likelihood algorithm were performed to assess the suitability of combining LiDAR intensity with digital images. These results were compared with those obtained using only information from the multispectral camera and showed an increase in accuracy of up to 40% in land-use discrimination after introducing normalized LiDAR intensity values. The combination of both data types allowed the classification of eight types of land use with an accuracy over 92%.