I am doing some work related to eye images.
I did edge detection to it. The edge is like a curve and not continuous. I have to assume it to be continuous and find normals to that curve. How do I find the normals to it using MATLAB?
you can see the image below.
I want to find the normals to the upper curve.
I hope that I was clear enough.
Even though it seems unintuitive, the edge direction at every pixel is a pretty good estimate of the normal. This would be the simplest solution, because it doesn't involve any curve fitting.
In MATLAB, you can find pixel-wise edge directions using the Sobel filter:
[BW,thresh,gv,gh] = edge(I,'sobel');
edgeDir = atan2(gv, gh);
This gives you the edge directions as angles in radians.
You may want to consider curve fitting (MSE based or some other criteria) to the data. I believe a second order will do good for the upper curve, and once you have a model you can can calculate the tangent and normal at each point.
As Zaphod recommended the normal is perpendicular to the edge. You don't need to do curve fitting, you can use back projection to identify the focal point of the curve.
Start at each edge point along the curve and draw a line from curve in the direction of the normal. Draw the line by incrementing the value of each pixel the line passes through. Once you do this for all the edges you would hope to find two pixels with higher values then the rest, one for each of your curves. You should then know by there locations which is the focal point for each curve.
Related
I have image of robot with yellow markers as shown
The yellow points shown are the markers. There are two cameras used to view placed at an offset of 90 degrees. The robot bends in between the cameras. The crude schematic of the setup can be referred.
https://i.stack.imgur.com/aVyDq.png
Using the two cameras I am able to get its 3d co-ordinates of the yellow markers. But, I need to find the 3d-co-oridnates of the central point of the robot as shown.
I need to find the 3d position of the red marker points which is inside the cylindrical robot. Firstly, is it even feasible? If yes, what is the method I can use to achieve this?
As a bonus, is there any literature where they find the 3d location of such internal points which I can refer to (I searched, but could not find anything similar to my ask).
I am welcome to a theoretical solution as well(as long as it assures to find the central point within a reasonable error), which I can later translate to code.
If you know the actual dimensions, or at least, shape (e.g. perfect circle) of the white bands, then yes, it is feasible and possible.
You need to do the following steps, which are quite non trivial to do, and I won't do them here:
Optional but extremely suggested: calibrate your camera, and
undistort it.
find the equation of the projection of a 3D circle into a 2D camera, for any given rotation. You can simplify this by assuming the white line will be completely horizontal. You want some function that takes the parameters that make a circle and a rotation.
Find all white bands in the image, segment them, and make them horizontal (rotate them)
Fit points in the corrected white circle to the equation in (1). That should give you the parameters of the circle in 3d (radious, angle), if you wrote the equation right.
Now that you have an analytic equation of the actual circle (equation from 1 with parameters from 3), you can map any point from this circle (e.g. its center) to the image location. Remember to uncorrect for the rotations in step 2.
This requires understanding of curve fitting, some geometric analytical maths, and decent code skills. Not trivial, but this will provide a solution that is highly accurate.
For an inaccurate solution:
Find end points of white circles
Make line connecting endpoints
Chose center as mid point of this line.
This will be inaccurate because: choosing end points will have more error than fitting an equation with all points, ignores cone shape of view of the camera, ignores geometry.
But it may be good enough for what you want.
I have been able to extract the midpoint by fitting an ellipse to the arc visible to the camera. The centroid of the ellipse is the required midpoint.
There will be wrong ellipses as well, which can be ignored. The steps to extract the ellipse were:
Extract the markers
Binarise and skeletonise
Fit ellipse to the arc (found a matlab function for this)
Get the centroid of the ellipse
hsv_img=rgb2hsv(im);
bin=new_hsv_img(:,:,3)>marker_th; %was chosen 0.35
%skeletonise
skel=bwskel(bin);
%use regionprops to get the pixelID list
stats=regionprops(skel,'all');
for i=1:numel(stats)
el = fit_ellipse(stats(i).PixelList(:,1),stats(i).PixelList(:,2));
ellipse_draw(el.a, el.b, -el.phi, el.X0_in, el.Y0_in, 'g');
The link for fit_ellipse function
Link for ellipse_draw function
Most of the times, I determine contour orientation generating 2D points and computing the closed polygon area. Depending on the area value sign I can understand if the contour is oriented clockwise or not (see How to determine if a list of polygon points are in clockwise order?).
Would it be possible to do the same computations without generating 2D points? I mean, relying only on geometric curve properties?
We are interested in determining the orientation of contours like these ones without sampling them with 2D points.
EDIT: Some interesting solutions can be found here:
https://math.stackexchange.com/questions/423718/general-way-to-find-out-whether-a-curve-is-positively-oriented
Scientific paper: Determining the orientation of closed planar curves, DJ Filip (1990)
How are those geometric curves defined?
Do you have an angle for them? The radius doesn't matter, only the difference between entry-angle and exit-angle of each curve.
In that case, a trivial idea crossing my mind is to just sum up all the angles. If the result is positive, you know you had more curves towards the right meaning it's a clockwise contour. If it was negative, then more curves were leftwards -> anti-clockwise contour. (assuming that positive angels determine a right-curve and vica versa)
After thinking about this for awhile, for polygons that contain arcs I think there are three ways to do this.
One, is to break the arcs into line segments and then use the area formula as described above. The success of this approach seems to be tied to how close the interpolation of the arcs is as this could cause the polygon to intersect itself.
A quicker way than the above would be to do the interpolation of the arcs and then find a vertex in the corner (minimal Y, if tie minimal X) and use the sign of the cross product for that vertex. Positive CCW, negative CW. Again, this is still tied to the accuracy of the interpolation.
I think a better approach would be to find the midpoint of the arc and create two line segments, one from the beginning of the arc to the midpoint and another from the midpoint to the end of the arc and replace the arc with these line segments. Now you have a polygon with only line segments. Then you can add up all the normalized cross products of all the vertices. The sign will tell you the direction. Positive is counter-clockwise, negative is clockwise. In this case it doesn't matter if the polygon self-intersects.
The picture below shows a triangular surface mesh. Its vertices are exactly on the surface of the original 3D object but the straight edges and faces have of course some geometric error where the original surface bends and I need some algorithm to estimate the smooth original surface.
Details: I have a height field of (a projectable part of) this surface (a 2.5D triangulation where each x,y pair has a unique height z) and I need to compute the height z of arbitrary x,y pairs. For example the z-value of the point in the image where the cursor points to.
If it was a 2D problem, I would use cubic splines but for surfaces I'm not sure what is the best solution.
As commented by #Darren what you need are patches.
It can be bi-linear patches or bi-quadratic or Coon's patches or other.
I have found no much reference doing a quick search but this links:
provide an overview: http://www.cs.cornell.edu/Courses/cs4620/2013fa/lectures/17surfaces.pdf
while this is more technical: https://www.doc.ic.ac.uk/~dfg/graphics/graphics2010/GraphicsHandout05.pdf
The concept is that you calculate splines along the edges (height function with respect to the straight edge segment itself) and then make a blending inside the surface delimited by the edges.
The patch os responsible for the blending meaning that inside any face you have an height which is a function of the point position coordinates inside the face and the values of the spline ssegments which are defined on the edges of the same face.
As per my knowledge it is quite easy to use this approach on a quadrilateral mesh (because it becomes easy to define on which edges sequence to do the splines) while I am not sure how to apply if you are forced to go for an actual triangulation.
As an input I have a segmented image of the upper body, I'm trying to detect shoulders from this image.
I minimized the region by a threshold calculated by simple known ratios between head size and shoulder width.
Now I have the shoulders region, performed edge detection on it.
Now I need to find the points of shoulders.
is there a fast way to detect the shoulder curves ?
I'm using Matlab.
This is my input image :
Bezier Curve is just a mathematical description of a curve, (linear interpolation, using control points).
It is not a curvetracer.
If you need bezier curve descriptions, you would need to do a best fit between a bezier curve model, and the data. Before you get started, you should probably play around with bezier curves, to get a feeling of how they operate.
See here: http://www.mathworks.com/matlabcentral/fileexchange/33828-generalised-bezier-curve-matlab-code
for a Bezier Curve render, in matlab.
It displays the bezier curve, when you provide some control points.
There is a few methods to actually fit a bezier curve to a set of data, here is one for matlab (using the least squares method).
http://www.mathworks.com/matlabcentral/fileexchange/15542-cubic-bezier-least-square-fitting
It will some times work nicely, and sometimes fail miserably, this is due to the least squares method, and the uniform parameterization used. It should work OK for your shoulder problem.
you need to extract the edge data, as data points, but that should be trivial
I have a multiple plants in a single binary image. How would I identify each leaf in the image assuming that each leaf is approximately elliptical?
example input: http://i.imgur.com/BwhLVmd.png
I was thinking a good place to start would be finding the tip of each leaf and then getting the center of each plant. Then I could fit the curves starting from the tip and then going to the center. I've been looking online and saw something involving a watershed method, but I do not know where to begin with that idea.
You should be aware that these things are tricky to get working robustly - there will always be a failure case.
This said, I think your idea is not bad.
You could start as follows:
Identify the boundary curve of each plant (i.e. pixels with both foreground and background in their neighbourhood).
Compute the centroid of each plant.
Convert each plant boundary to a polar coordinate system, with the centroid as the origin. This amounts to setting up a coordinate system with the distance of each boundary curve point on the Y axis and the angle on the X axis.
In this representation of the boundary curve, try to identify maxima; these are the tips of the leaves. You will probably need to do some smoothing. Use the parts of the curve before and after the maxima the start fitting your ellipses or some other shape.
Generally, a polar coordinate system is always useful for analysing stuff thats roughly circular.
To fit you ellipses, once you have a rough initial position, I would probably try an EM-style approach.
I would do something like this (I is your binary image)
I=bwmorph(bwmorph(I, 'bridge'), 'clean');
SK=bwmorph(I, 'skel', Inf);
endpts = bwmorph(SK,'endpoints');
props=regionprops(I, 'All');
And then connect every segment from the centroids listed in props.centroid to the elements of endpts that should give you your leaves (petals?).
A bit of filtering is probably necessary, bwmorph is your friend. Have fun!