I am developing a project of detecting vehicles' headlights in night scene. I am working on a demo on MATLAB. My problem is that I need to find region of interest (ROI) to get low computing requirement. I have researched in many papers and they just use a fixed ROI like this one, the upper part is ignored and the bottom is used to analysed later.
However, if the camera is not stable, I think this approach is inappropriate. I want to find a more flexible one, which alternates in each frame. My experiments images are shown here:
If anyone has any idea, plz give me some suggestions.
I would turn the problem around and say that we are looking for headlights
ABOVE a certain line rather than saying that the headlights are below a certain line i.e. the horizon,
Your images have a very high reflection onto the tarmac and we can use that to our advantage. We know that the maximum amount of light in the image is somewhere around the reflection and headlights. We therefore look for the row with the maximum light and use that as our floor. Then look for headlights above this floor.
The idea here is that we look at the profile of the intensities on a row-by-row basis and finding the row with the maximum value.
This will only work with dark images (i.e. night) and where the reflection of the headlights onto the tarmac is large.
It will NOT work with images taking in daylight.
I have written this in Python and OpenCV but I'm sure you can translate it to a language of your choice.
import matplotlib.pylab as pl
import cv2
# Load the image
im = cv2.imread('headlights_at_night2.jpg')
# Convert to grey.
grey_image = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
Smooth the image heavily to mask out any local peaks or valleys
We are trying to smooth the headlights and the reflection so that there will be a nice peak. Ideally, the headlights and the reflection would merge into one area
grey_image = cv2.blur(grey_image, (15,15))
Sum the intensities row-by-row
intensity_profile = []
for r in range(0, grey_image.shape[0]):
intensity_profile.append(pl.sum(grey_image[r,:]))
Smooth the profile and convert it to a numpy array for easy handling of the data
window = 10
weights = pl.repeat(1.0, window)/window
profile = pl.convolve(pl.asarray(intensity_profile), weights, 'same')
Find the maximum value of the profile. That represents the y coordinate of the headlights and the reflection area. The heat map on the left show you the distribution. The right graph shows you the total intensity value per row.
We can clearly see that the sum of the intensities has a peak.The y-coordinate is 371 and indicated by a red dot in the heat map and a red dashed line in the graph.
max_value = profile.max()
max_value_location = pl.where(profile==max_value)[0]
horizon = max_value_location
The blue curve in the right-most figure represents the variable profile
The row where we find the maximum value is our floor. We then know that the headlights are above that line. We also know that most of the upper part of the image will be that of the sky and therefore dark.
I display the result below.
I know that the line in both images are on almost the same coordinates but I think that is just a coincidence.
You may try downsampling the image.
Related
I have several images that i would like to correct from artifacts. They show different animals but they appear to look like they were folded (look at the image attached). The folds are straight and they go through the wings as well, they are just hard to see but they are there. I would like to remove the folds but at the same time preserve the information from the picture (structure and color of the wings).
I am using MATLAB right now and i have tried several methods but nothing seems to work.
Initially i tried to see if i can see anything by using an FFT but i do not see a structure in the spectrum that i can remove. I tried to use several edge detection methods (like Sobel, etc) but the problem is that the edge detection always finds the edges of the wings (because they are stronger)
rather than the straight lines. I was wondering if anyone has any ideas about how to proceed with this problem? I am not attaching any code because none of the methods i have tried (and described) are working.
Thank you for the help in advance.
I'll leave this bit here for anyone that knows how to erase those lines without affecting the quality of the image:
a = imread('https://i.stack.imgur.com/WpFAA.jpg');
b = abs(diff(a,1,2));
b = max(b,[],3);
c = imerode(b,strel('rectangle',[200,1]));
I think you should use a 2-dimensional Fast Fourier Transform
It might be easier to first use GIMP / Photoshop if a filter can resolve it.
I'm guessing the CC sensor got broken (it looks to good for old scanner problems). Maybe an electric distortion while it was reading the camera sensor. Such signals in theory have a repeating nature.
I dont think this was caused by a wrong colordepth/colorspace translation
If you like to code, then you might also write a custom pixel based filter in which you take x vertical pixels (say 20 or so) compare them to the next vertical row of 20 pixels. Compare against HSL (L lightnes), not RGB.
From all pixels calculate brightness changes this way.
Then per pixel check H (heu) is within range of nearby pixels take slope average of their brightness(ea take 30 pixels horizontal, calculate average brightnes of first 10 and last 10 pixels apply that brightness to center pixel 15,... //30, 15, 10 try to find what works well
Since you have whole strokes that apear brighter/darker such filter would smooth that effect out, the difficulty is to remain other patterns (the wings are less distorted), knowing what color space the sensor had might allow for a better decision as HSL, maybe HSV or so..
i want to get the red region as specified in the image below :
remember that the red region that is shown in the image is just for clarification , it is not present in original image , below is the original image attached :
i also have the iris point in this region, i already got that point , if that point can help me so i can share that image too.
can someone help me in this .....
For this specific image, let's call it BW, you can find the center region as:
BWnoBorder= imclearborder(BW); %# remove the white that touches the border
OnlyCenter = bwareaopen(BWnoBorder,1000); %# remove all small pixel areas
A robust method might be the snake region growing algorithm.
Seems like you thresholded the eye illuminated by IR or something. To answer your question or even to ask it correctly you have to show a number of images to evaluate the stability and noise in eye socket regions. Otherwise one can come up with a solution that works for the image above but not in general.
For example, I can invert your image, get the largest Connected Component (the dark region) and erode it till it becomes thin, see below. It is easy to get an ellipse from this binary mask but will it work in general case with your noisy input?
A good thing to start is to say what do you expect to find. Say you are looking for an eye, dark surrounding area and a bright skin tone - make it 3 Mixed models that settle simultaneously in the EM fashion. Provide some shape priors to increase accuracy. think about other visual cues such as specs on the iris, saccades, FA from blinks, etc.
I'm trying to consistently detect a certain color between images of the same scene. The idea is to recognize a set of object based on a color profile. So, for instance, if I'm given a scene with a green ball in it and I select that green as part of my color palette, I would like a function which has a matrix reflecting that it detects the ball.
Can anyone recommend some matlab functions/plugins/starting points for this project? ideally the function for color recognition will take an array of color values and will match them within a certain threshold.
Kinda like this:
http://www.mathworks.com/matlabcentral/fileexchange/18440-color-detection-using-hsv-color-space-training-and-testing
except it works (this one didn't)
Update:
Here's why I chose not to use the above toolkit..
I start by selecting some colors of interest in the picture
and then ask the function to recognize the road in later images...
And absolutely nothing useful is triggered. So yeah, apart from the few bugs that I came across in the code on download and fixed, this was kind of the kicker. I didn't try to fix the body of the code that recognizes the colors because.. well, I don't know how, which is why I came here.
So, let me just start off by saying road detection with color profiles is a pathological problem. But if the color of the roads are consistent, and the lighting doesn't change the color of the object you are trying to recognize then you might have a shot. (this will be extremely difficult if this is taken outside, or with different cameras, or if shadows happen, or it taking place in any sort of real-world environment)
Here are a few things that might help.
Try smoothing the image beforehand, the reason you get the bad results in the first images is probably because of small pixel variations in the road. If you can blur them, or use some sort of watershed or local averaging, you might get regions with more consistent color.
You might also consider using the LAB color space instead of HSV or RGB.
Using edge detection (see matlab's canny edge detector) might be able to get you some boundary information. If you are looking for a smooth object, there will not be very many edges in it.
Edit: I tried to adhere to this advice in the most simplistic way. Here are the resulting code and a few samples.
im=rgb2gray(im) %for most basic color capturing.. using another color space is better practice
%imshow(im)
RoadMask=roipoly(im)%create mask
RoadMask=uint8(RoadMask);%cast to so you can elementwise multiply
im=im.*RoadMask;%apply mask
[x y]=size(im);
for i=1:x
for j=1:y
%disp('here')
if (im(i,j)<160 || im(i,j)>180) %select your values based on your targets range
im(i,j)=0; %replace everything outside of range with 0
%disp(im(x,y)) %if you'd like to count pixels, turn all values
end %within range to 1 and do a sum at end
end
end
First converted from RGB to grayscale
selected a region that generally matched the roads grayness
Notice parts of the road are not captured and the blocky edges. such as this -------------^
This implementation was quicky and dirty, but I wanted to put it up before I forgot. I'll try to update with code that implements smoothing, sampling, and the LAB color space.
I have an image which looks like this:
I have a task in which I should circle all the bottles around their opening. I created a simple algorithm and started working it. My algorithm follows:
Threshold the original image
Do some morphological opening in it
Fill the empty holes
Separate the portion of the image using region props such that only the area equivalent to the mouth of the bottles is selected.
Find the centroid for each and draw circle around each bottle.
I did according to the algorithm above and but I have some portion of the image around which I draw a circle. This is because I have selected the area since the area of the mouth of bottle and the remained noise is almost same. And so I yielded a figure like this.
The processing applied on the image look like this:
And my final image after plotting the circle over the original image is like this:
I think I can deal with the extra circle, that is, because of some white portion of the image remained as shown in the figure 2 below. This can be filtered out using regionproping for eccentricity. Is that a good idea or there are some other approaches to this? How would I deal with other bottles behind the glass and select them?
Nice example images you provide for your question!
One thing you can use to detect the remaining bottles (if there are any) is the well defined structure of the placement of the bottles.
The 4 by 5 grid of the bottle should be relatively easy to locate, and when the grid is located you can test if a bottle is detected at each expected bottle location.
With respect to the extra detected bottle, you can use shape features like
eccentricity,
the first Hu moment
a ratio between the perimeter length squared over the area (which is minimized for a circle) details here
If you are able to detect the grid, it should be easy to located it as an outlier (far from an expected bottle location) and discard accordingly.
Good luck with your project!
I've used the same approach as midtiby's third suggestion using the ratio between area and perimeter called shape factor:
4π * Area /perimeter^2
to detect circles from a contour traced image (from the thresholded image) to great success;
http://www.empix.com/NE%20HELP/functions/glossary/morphometric_param.htm
Regarding the 4 unfound bottles, this is rather tricky without some a priori knowledge of what it is you're looking at (as discussed using the 4 x 5 grid, then looking from the centre of each cell). I did think that from the list of contours, most would be of the bottle tops (which you can test using the shape factor stuff), however, one would be of a large rectangle. If you could find the extremities of the rectangle (from the largest contour in terms of area), then remove it from the third image, you'd be left with partial circles. If you then contour traced those partial circles and used a mixture of shape factor/curve detection etc. may help? And yes, good luck again!
I recall seeing a paper a while back for an algorithm that could automatically and seamlessly "graft" texture from parts of an image onto another part of an image.
The approach was something along the lines of the following:
You'd build up a databases of small squares of pixels (perhaps 8X8) from the parts of the picture that are present.
You'd then pick an empty pixel (the "destination" for the texture graft) to fill in, and look for one of the squares in your database that most closely matches the surrounding pixels. You'd then color the empty pixel according to the color of the corresponding pixel in the square you find. Then you pick another empty pixel and repeat until there are no empty pixels remaining.
Of course, this is only a vague description because I can't find any references to this algorithm to refresh my memory of the details! Can anyone help?
Sounds a lot like Texture Synthesis by Non-parametric Sampling