This section describes each button on the edge tool

**im rank**Pre-process the images to replace each greylevel value with the local rank (within a region typically 11x11 pixels). This process simulates the first order effect of change detection found in the primate retina in a way which produces data with approximately binomial errors. This is sometimes more suitable for input to correlation based stereo matching than raw greylevels.**im rectify**Transform both the left and right stereo images to cyclopean co-ordinates. Don't expect the images to still be centered in the*left*and*right**Tv*s.**canny**Linked canny edges are obtained, to sub-pixel accuracy, for the selected region of interest. The process is two pass (implementing hysteresis); first the edges with an intensity gradient above a low threshold are identified; then linked edge strings that are entirely below an upper threshold are eliminated ( the remaining edges are displayed in purple with blue termination points).**Edge paramaters:**Sigma The parameter of the gaussian convolution profile used to smooth the image prior to the gradient calculation. It essentially selects the scale of the edge detector. Precision The ratio of the smallest stored value of the guassian convolution profile with respect to the largest. Lower values give increased precision, larger convolution masks and increased computation time. Lower thres The lower contrast threshold (see above discussion). Upper thres The upper contrast threshold (see above discussion). Length thres A lower threshold on the lenght of connected edge strings. Shorter sstrings are eliminated.

**edge rectify**Use current parallel camera geometry ( recovered from the calibration files) to update the position of edge data to be consistent with it. The actual storage location of each image is unaffected. This process must be performed prior to stereo matching and 3D geometrical recovery.**edge derect**Reverses the effect of "edge rectify" to return position to the initial coordinate frame.**geom2**Performs geometrical approximation procedures over each of the linked edges strings identified in the left image.**Geom2 Options :**2D Geom Fit poly: standard polygonal approximation algorithm. linear: recursively identifies the current longest straight section, segments it, and applies itself to the remaining sections. The actual line fit is done by orthogonal regression. Sections between straight lines found by "linear" are fit by "poly". conic: recursively identifies the current longest conic section, segments it, and applies itself to the remaining sections. The actual conic fit is done using the bias corrected Kalman filter. Sections between conics are fit by linear. All conics are coerced to be ellipses. linear/conic: [default] first applies linear, then emplys a few heuristics to identify groups of neighbouring straight sections that could be conics and applies conic to underlying edge string. Any conics that are found are extended if possible into neighbouring straight lines and/or conic sections (each new section must be absorbed as a whole). This results in unique descriptions for each edge string as either a straight line or conic section. 2D Options join: after initial fit, try to combine conic sections with other conics and/or straight lines that were not necessariliy neighbouring on, or even from, the same underlying edge string.

**Geom2 Parameters :**Low fit thres Lower threshold or primitive fitting in pixels. Can be sustained for short deviations of edge strings from primitive. Up fit thres Upper threshold of primitive fitting in pixels. Can not be violated.