Detecting Greenery in Near Infrared Images of Ground-level Scenes

Transcription

1 Detecting Greenery in Near Infrared Images of Ground-level Scenes Piotr Łabędź Agnieszka Ozimek Institute of Computer Science Cracow University of Technology

2 Digital Landscape Architecture, Dessau Bernburg Problems concerning ground-level views

3 Three components of the image: man-made substance (building, infrastructure), natural substance (plants, greenery), the background.

4 Automatic background detection

5 Colour-based sky characteristics

6 Intensity in RGB colour channels Value 150 R G B sample number Colour-based sky characteristics

7 A binary image with the man-made objects marked white

8 Colour similarity between vegetation and man-made objects

9 Spectral characteristics of leaves

10 Normalized Difference Vegetation Index (NDVI) NDVI= NIR NIR + R R where: NIR near infrared channel R red channel

11 Enhanced Vegetation Index (EVI) EVI= G NIR NIR R + C R + C B L where: NIR/R/B colour channels: near infrared, red and blue, respectively, L - the canopy background, C 1 and C 2 coefficients, considering aerosol resistance in the atmosphere, G gain factor, L soil adjusted factor.

12 MODIS EVI (Moderate Resolution Imaging Spectroradiometer) MODIS EVI= 2,5 NIR + NIR R 6 R + 7,5 * B + 1

13 1.2 SUNLIGHT Relative sensitivity CCD HUMAN VISION wavelenght [nm] UV IR (infrared)

14 Reading an image in the visible light Red channel separation Reading an image in infrared NIR NDVI calculation Numerator calculation = difference NIR - R Denominator calculation = sum NIR + R Quotient calculation numerator / denominator Histogram calculation Binarization using Otsu algorithm Y Image closing Resultant image writing Is the result satisfying? N Manual choice of threshold An algorithm of greenery detection basing on NDVI

15 Reading an image in the visible light Reading an image in infrared NIR Red and blue channels separation Numerator calculation = difference NIR - R Modis EVI calculation Denominator calculation = sum NIR + 6.5*R + 7*B + 1 Calculation of the equation: 2.5*(numerator / denominator) Histogram calculation T Binarization using Otsu algorithm Y Is the result satisfying? N Image closing Resultant image writing Manual choice of threshold An algorithm of greenery detection basing on EVI

16 An example of a photograph in the visible light

17 Blue and green channels

18 Red and infrared channels

19 NDVI versus MODIS EVI

20 EVI without calibrating coefficients and the difference between it and NDVI

21 The background and greenery filtered out

22 Various lighting conditions

23 Various lighting conditions

24 Various lighting conditions a difference in plants detection

25 Man-made objects detected using the algorithm proposed

26 Variations in the colour of vegetation with changes in season

27 Greenery detection

28 Cultural components of the analysed scene

29 Advantages of the method of automatic greenery detection: Effective distinction between the verdure and green man-made objects. Clear distinction of plants parts located in the shade - high level of reflectance in near infrared, and low in the red channel. Correct results in distant parts of the view. Effectiveness of Otsu algorithm of binarization.

30 Errors occurring in the resultant images: Tree trunks and branches, dry grass - the absence of chlorophyll. The other natural elements (soil, rocks, water) - detection impossible, (the lack of the chlorophyll). Yellow leaves increasing reflectance in the red channel. Inaccuracies in the objects contours, - the consequence of pixels values interpolation (antialiasing). Reflections in the glossy surfaces spectral similarity with plants. Photographs taken against the light, the branches not fully covered with leaves.