| Summary: | 碩士 === 國立臺北科技大學 === 電機工程研究所 === 104 === In Taiwan, the forests distribute vertically along the central region and can be categorized into hardwood, mixed, and conifer forests. Terrain features in mountainous areas make manual inspection of forests very difficult if not impossible. By utilizing remote sensing techniques the efforts of field sampling could be effectively reduced.
The related remote sensing research was beginning from digital imagery back to the mid-1980s. One of the earliest work of such algorithms presented by Pinz proposed the vision expert system to locate the center of a crown and estimate the crown radius by searching for local brightness maxima in smoothed aerial images. Afterwards many algorithms had been proposed and developed. In recent years, Light Detection And Ranging (LiDAR) data have emerged as sources for estimating tree height, stem volume and biomass at stand level due to strengths of three dimensional structures. High sampling LiDAR data provides detailed vertical structure of tree crowns. Although some algorithms developed for high-resolution optical imagery could be applied for extracting individual tree information from LiDAR data, few of them were specifically designed to utilize 3D properties provided by LiDAR data.
The template matching algorithms shed light on ideas of the algorithm proposed in this thesis. The templates are constructed by considering a three-dimensional description of individual tree-crown envelope. Pollock proposed a well-known model to construct a synthetic comprehensive image template considering both geometric and radiometric characteristics of individual tree. The Pollock model helps to increase detection rate if parameters of models were set properly. Nevertheless, it is always challenging to determine parameters since each individual tree has different grown condition. To address this problem, priori knowledge about the scene is required to predetermine an appropriate range of parameters for the models.
This thesis proposed an iterative algorithm, called Least Square Fitting of Pollock Model. It designs an iterative process to look for the Pollock model best fitting the real 3D structure of each extracted tree top. The proposed algorithm provide a solution to avoid difficulties of finding the best set of parameters for each tree crown. As a result, it improves the detection rate and decreases the computing time as shown in the experimental studies. The proposed algorithm is more suitable for practical applications.
|
|---|
