Feature Tracking

Description

The tool performs displacement analysis using image cross-correlation between two raster datasets. For each raster cell, a reference window and a search window are defined. The reference window extracts a template pattern, which is then matched within the larger search window. The algorithm identifies the correlation peak and uses its position to calculate the magnitude and direction of displacement. Two correlation methods are available: local normalized cross-correlation (NCC) and an FFT-based NCC approach. The latter is ported from the original Fortran implementation of the IMCORR software developed by the National Snow and Ice Data Center. The FFT-based method computes correlations for all possible displacements simultaneously in the frequency domain, making it computationally efficient, particularly for large search areas. However, because the normalization is not strictly evaluated for each local window, the result is an approximation of true NCC. In contrast, the local NCC method evaluates the correlation independently for each candidate displacement using window-specific statistics (mean and variance). It supports different kernel shapes and optional Gaussian weighting to reduce edge effects within the matching window. Although this approach is computationally more expensive, it provides greater robustness to local intensity variations and typically yields higher accuracy. The maximum displacement of a feature that can be measured is: (search chip size/2 - reference chip size/2) * cell size This maximum applies only to exactly diagonal motion. In order to measure larger distances, a known offset between reference and search window can be given. The offset specifies the x/y components of the expected movement. Besides a constant offset, spatially variable offsets can be applied when a point shapes layer with start positions and x- and y-displacements, e.g. from a previous tool run, is provided. The tool can take two elevation models to query the elevations at the start and end points of the each detected movement to compute a 3D displacement vector. The tool computes the following attributes (some of them are only available if elevation data is provided):

• ID: a unique identifier per measurement
• GX, GY: grid column and row indices of the source (start) cell
• REALX, REALY, REALZ: x,y,z world coordinates of the source (start) cell of the movement
• DISP: horizontal displacement distance [map units]
• DISP_REAL: 3D displacement distance [map units]
• STRENGTH: correlation strength of the Fourier analysis [-]
• FLAG: flag indicating a correlation was found [-]
• XDISP, YDISP, ZDISP: x,y,z components of the displacement [map units]
• XDISP_UNIT, YDISP_UNIT: x,y components of the displacement unit vector [0-1]
• XTARG, YTARG, ZTARG: x,y,z world coordinates of the target (stop) cell of the movement
• XERR, YERR: x,y error components of the correlation strength [-]
• ASPECT, SLOPE: compass direction and gradient of the displacement [degree]
• ID_M: identifier of the spatially variable measurement position (if specified)

References

Fahnestock, M. A., Scambos, T. A., and Bindschadler, R. A. (1992): Semi-automated ice velocity determination from satellite imagery. Eos, 73, 493.

Fey, C., Rutzinger, M., Wichmann, V., Prager, C., Bremer, M., and Zangerl, C. (2015): Deriving 3D displacement vectors from multi-temporal airborne laser scanning data for landslide activity analyses. GIScience & Remote Sensing, 52(4), 437-461. Link

Scambos, T. A., Dutkiewicz, M. J., Wilson, J. C., and Bindschadler, R. A. (1992): Application of image cross-correlation to the measurement of glacier velocity using satellite image data. Remote Sensing Environ., 42(3), 177-186. Link

Parameters

Command Line

Usage: saga_cmd grid_analysis 19 [-GRID_1 ] [-GRID_2 ] [-DTM_1 ] [-DTM_2 ] [-CORRPOINTS ] [-CORRLINES ] [-CORR_METHOD ] [-SEARCH_CHIPSIZE ] [-REF_CHIPSIZE ] [-KERNEL_TYPE ] [-WEIGHTING ] [-STEPPING ] [-OFFSET_TYPE ] [-DISP_X ] [-DISP_Y ] [-SHP_OFFSET ] [-FIELD_DISP_X ] [-FIELD_DISP_Y ] [-FIELD_ID ] [-MAX_DIST ]
  -GRID_1:            Grid 1
	grid, input
  -GRID_2:            Grid 2
	grid, input
  -DTM_1:             DTM 1
	grid, input, optional
  -DTM_2:             DTM 2
	grid, input, optional
  -CORRPOINTS:        Correlated Points
	shapes, output
  -CORRLINES:         Displacement Vector
	shapes, output
  -CORR_METHOD:       Correlation Method
	choice
	Available Choices:
	[0] local NCC
	[1] FFT-based NCC
	Default: 0
  -SEARCH_CHIPSIZE:   Search Chip Size
	choice
	Available Choices:
	[0] 32x32
	[1] 64x64
	[2] 128x128
	[3] 256x256
	Default: 1
  -REF_CHIPSIZE:      Reference Chip Size
	choice
	Available Choices:
	[0] 16x16
	[1] 32x32
	[2] 64x64
	[3] 128x128
	Default: 1
  -KERNEL_TYPE:       Kernel Type
	choice
	Available Choices:
	[0] Square
	[1] Circle
	Default: 1
  -WEIGHTING:         Weighting
	choice
	Available Choices:
	[0] no weighting
	[1] Gaussian weighting
	Default: 1
  -STEPPING:          Stepping
	integer number
	Minimum: 1
	Default: 2
  -OFFSET_TYPE:       Offset Type
	choice
	Available Choices:
	[0] no offset
	[1] constant offset
	[2] spatially variable offset
	Default: 0
  -DISP_X:         X Displacement
	floating point number
	Default: 0.000000
  -DISP_Y:         Y Displacement
	floating point number
	Default: 0.000000
  -SHP_OFFSET:        Displacement Offsets
	shapes, input
  -FIELD_DISP_X:      X Displacement
	table field
  -FIELD_DISP_Y:      Y Displacement
	table field
  -FIELD_ID:          ID
	table field
  -MAX_DIST:       Maximum Distance
	floating point number
	Minimum: 0.000000
	Default: 4.000000