AUTOMATED CITRUS TREE COUNTING FROM OBLIQUE OR ORTHO IMAGES AND TREE HEIGHT ESTIMATION FROM OBLIQUE IMAGES

Abstract

The objectives of this research paper were to 1) compare tree counts between oblique and ortho images and 2) determine the accuracy of tree height measurements from oblique images. Accurate determination of tree count and canopy volume can aid growers in planning some orchard operations and estimating yield. The analytical capability of Feature Analyst (FA) to count trees enables the user to quantify the trees in a given area. Eight sample sets of ortho and oblique images were selected from Pictometry aerial imagery software (Pictometry International Corp, 100 Town Center Drive, Suite A, Rochester, NY, 14623) by entering the coordinates of the required area of interest. The images thus obtained were exported to Arc Map in ortho and oblique views. The same training samples randomly selected from the study area in the ortho view were also used for the oblique image in FA to validate/determine the number of trees. The image was represented by a shape file, and an attribute query denoted the quantity of trees. After processing with FA, the majority of the machine counts were below the actual value. Three of the eight ortho images represented over counts, while five represented under count with a total range of -16 to +23%, with a mean error of 8.67%. All the eight oblique images represented under counts with an error range of 0.26 to 27.82% with a mean error of 9.94%. Thus, Oblique images had an error range of 27.56%, while orthogonal images have an error range of 39%, but overall the average error was less for ortho images. An estimate was made of tree height using oblique images in Pictometry. The errors were partially due to the ambiguity in selecting each tree’s base and apex. Assuming there may be 0.15 m error due to resolution in ground area covered and tree height centering contributed an 11% error, overall precision in tree canopy volume measurement was still 85%. INTRODUCTION Feature Analyst (FA) has the ability to count trees in a given acreage. Saraswat et al. (2007) used FA to count the trees in citrus groves in Polk County, Florida. They digitized a few trees in the study area and used these samples to determine the total number of trees in the study area (Saraswat et al., 2008). Attempts were made by previous researchers to determine tree height using Lidar data and digital roof models, but heights could not be determined to 100% accuracy. Imai et al. created a digital height model (DHM) by subtracting Digital Elevation model from a Digital Surface model. Hashiba et al. (2003) generated a digital roof model (DRM) using a typical stereogram concept and basic trigonometric calculations were executed to calculate the tree height. Pictometry is aerial imaging software that has the capabilities to measure the tree height and ground area covered from ortho and oblique images. The aerial images were collected by running flight line mosaics across the study area. These two factors are required to calculate the tree canopy volume. The combination of these factors to determine canopy volume per block then enables estimating the yield with weather factors that determine flowering intensity (Valiente and Albrigo, 2004; Albrigo et al., 2002). Proc. IS on Appl. of Precision Agric. for Fruits and Vegetables Eds.: L.G. Albrigo and R. Ehsani Acta Hort. 824, ISHS 2009 92 MATERIAL The Citrus Research and Education Center groves in Florida were chosen as the study area. The images were randomly selected from four blocks. Eight other sites were identified and ortho and oblique images were obtained for these sites. These were 0.15 m resolution images provided by the Pictometry Visual Intelligence Company. Images were collected by running flight line mosaics across the study area. Each image had three bands (RGB) stacked as a true color composite. Five additional sites were identified and ortho images were extracted from the Pictometry image source. The same site areas (0.3 m resolution) were extracted from a Polk county tax assessor source. All the trees in the study area were citrus.