I am new on computer vision, I have a project that can segments floor after that it can change floor. I have good segmentation model but I don't know how to change pattern properly. As an example if somebody uploads this picture;
After that, if gives this image as input;
output should be like that;
As you can see in the image changed floor rotated and well fitted to room.
It can be any of room picture but pattern always fixed like an example that I gived above. I can do converting and integration operations but I don't know how to get room rotation and and camera angle automatically. I am open to any kind of suggestions or resources to topic in computer vision about for thath kind of operations.
How can I solve this?
Note: Sorry about my poor english. I hope I could manage to describe my problem.
Related
I made a tif image based on a 3d model of a woodsheet. (x, y, z) represents a point in a 3d space. I simply map (x, y) to a pixel position in the image and (z) to the greyscale value of that pixel. It worked as I have imagined. Then I ran into a low-resolution problem when I tried to print it. The tif image would get pixilated badly as soon as it zooms out. My research suggests that I need to increase the resolution of the image. So I tried a few super-resolution algos found from online sources, including this one https://learnopencv.com/super-resolution-in-opencv/
The final image did get a lot bigger in resolution (10+ times larger in either dimension) but the same problem persists - it gets pixilated as soon as it zooms out, just about the same as the original image.
Looks like quality of an image has something to do not only with resolution of it but also something else. When I say quality of image, I mean how clear the wood texture is in the image. And when I enlarge it, how sharp/clear the texture remains in the image. Can anyone shed some light on this? Thank you.
original tif
The algo generated tif is too large to be included here (32M)
Gigapixel enhanced tif
Update - Here is a recently achieved result: with a GAN-based solution
It has restored/invented some of the wood grain details. But the models need to be retrained.
In short, it is possible to do this via deep learning reconstruction like the Super Resolution package you referred to, but you should understand what something like this is trying to do and whether it is fit for purpose.
Generic algorithms like the Super Resolution is trained on variety of images to "guess" at details that is not present in the original image, typically using generative training methods like using the low vs high resolution version of the same image as training data.
Using a contrived example, let's say you are trying to up-res a picture of someone's face (CSI Zoom-and-Enhance style!). From the algorithm's perspective, if a black circle is always present inside a white blob of a certain shape (i.e. a pupil in an eye), then next time it the algorithm sees the same shape it will guess that there should be a black circle and fill in a black pupil. However, this does not mean that there is details in the original photo that suggests a black pupil.
In your case, you are trying to do a very specific type of up-resing, and algorithms trained on generic data will probably not be good for this type of work. It will be trying to "guess" what detail should be entered, but based on a very generic and diverse set of source data.
If this is a long-term project, you should look to train your algorithm on your specific use-case, which will definitely yield much better results. Otherwise, simple algorithms like smoothing will help make your image less "blocky", but it will not be able to "guess" details that aren't present.
paper "Fast Obstacle Detection Using U-Disparity Maps with Stereo Vision"
reference paper: "Fast Obstacle Detection Using U-Disparity Maps with Stereo Vision"
I want to ask can opencv do this reslut?
How?
Thank you
Thanks for #Liew Xun's comment.
Input are left image and right image from two camera.(stereo camera)
I want to use opencv's cv2.StereoBM to do the result like the paper's image.
But,unfortunately i failed. I can't make the result like this paper's image.
So i am here to ask does anybody have the same experience do disparity map.
Can you give me a help or give me some advice?
What you are asking is a mobile robotics visual perception related question.
I guess this author made a trick on you as it is not called U disparity
The stereo depth can be computed by opencv stereo matching. If you fail, there could be many reason. you need to post what you have. you can follow this link to get depth map. https://docs.opencv.org/master/dd/d53/tutorial_py_depthmap.html
This right image is done by concatenating depth observation along the vertical plane. It is called V disparity. For normal road scene, most of the background road should be on the same straight line spam diagonally. Anything not on this line can be considered as an obstacle above or below the road surface. When the sensing distance is far, there might be deformation for the line because of stereo uncertainty. This algorithm is designed to detect cases such as negative obstacle(drains, potholes) and positive obstacle( cars humans）
you can find a sample implementation here
https://github.com/thinkbell/v_disparity
If you are Phd candidate, you need to read a lot more paper to know whether some paper is lying on their contribution rather than post question here and hope someone can tell you the answer.
Regards
Dr Yuan Shenghai
so I'm really new here. Currently working on a public art project where I need a little help with the programming because I'm kind off lost between codes.
First I'll give you a short description of the goal of the work and then state my problem.
I'm putting a webcam in the shopwindow of a gallery that is facing out on a public street. This webcam is connected to a tv screen that is facing outwards on the street so people see themselves being filmed (like cctv). Then if people stand still long enough for the camera the webcam makes an automatic screenshot what will be emailed to a site which hold a script for automatic attachment printing and the people from the street instantly come in to my gallery, on paper.
(and yes I have permission from the gallery to do this since it is slightly in the grey area of legality)
I come from a art background with interest in programming so this was all very very new for me and made it already quite far I think. I have a raspberry pi running with open cv and put a script on it for deep learning object detection (https://www.pyimagesearch.com/2017/09/18/real-time-object-detection-with-deep-learning-and-opencv/) < the link I used for that.
I also come across loads of pedestrian tracking but did not find a suitable code yet for a real time video stream.
So what I need from you guys, is a little help with how to make a timer in the script so that when people stand still long enough for the camera, it wil make the screenshot. It is a bit like reversed security cams script because they react on movement and I want it to react to no movement at all.
The automatic attachment printing part I got covered I think because there are a lot of scripts already on the internet.
If you have any tips or tricks.. please let me know.
Help a girl out!
Marije
There are a number of things you can try.
Is the camera faced towards a shopping street? In that case you could go for simple background subtraction. For each frame, apply some preprocessing (e.g. blurring, morpholoy operations), call findContours and compute the center of minEnclosingRect for each of these.
Another option is to use the inbuilt (and pretrained) HOG PeopleDetector. This is based on SVM (Support Vector Machines), which is another machine learning technique. For this to work efficiently you'd have to tune the parameters adequately. Since you're using a Pi you'd also need to consider the tradeoff between speed and accuracy. Using this technique, we'd be left with rectangles as well, so we can again compute the center.
For both techniques, you'd want to make sure that the center point doesn't fluctuate too much from frame to frame (that would mean the person is moving). For this you'd also want to take into account the framerate and understand that you can't guarantee person detection for every frame.
The caveat of the first technique, whilst having more explanatory power, would be that it'd detect ANYTHING that changes from frame to frame, that includes pets, bikes, cars (if on a public street) and so on. You could then consider filtering (e.g. by area, color).
recently I have been playing with the 360 fly HD camera and wondering if Aruco Marker can be detected during real time. The first thing come to my mind is to convert the fisheye image into perspective image first and then perform the detection on the perspective image(I am gonna try it and will update my result here later).
Converting a fisheye image into a panoramic, spherical or perspective projection
Hugin HowTo: Convert 360 Image to Cropped Flat Panoramic Image
I am not an expert in this field. Has anyone done this before? Is this something can be achieved by calibrating the camera differently such as correcting the camera matrix and distortion coefficient matrix?
If I am heading to the wrong direction, please let me know.
I was able to get a better understanding during the process.
First, I want to say that 360(fisheye, spherical, however you call it) image is NOT distorted. I was so tricked by my intuition and thought that the image was distorted based on what it looks like. NO it is not distorted. Please read enter link description here for more information.
Next, I have tried both 360 fly cameras and neither works. Every time I tried to access the camera with opencv, it automatically powers off and switch to storage mode. I guess the 360 dev team purposely implements this switching function to prevent "hacking" of their products. But, I've seen people successfully hacked the 360 fly, it's definitely workable.
At last, I was able to detect Aruco with Ricoh theta V(theta S should also work). It's so developer friendly and I was able to make it run in my first attempt. You just have to select the right camera and let the code run. The only problem is the range, which is expected(about 6ft) and Ricoh camera is kind of expensive($499).
click here to view succesful detection
I am writing a simple fly tracking software and I would love some input from opencv experts.
The image I have looks pretty much like:
I used to do tracking using kmeans and PIL/numpy but I re-wrote everything to use blob detection in opencv. Tracking works OK but I would also like to automatize division of ROI.
What I need to do is find each of the 32 grooves that appear in the picture, where flies live. See the black rectangle on the image as example of what I mean.
I think cornerHarris may be what I need but how do I specify only the grooves and not each single rectangle found in the image? All those grooves have proportions of roughly 10:1.
Thanks!
I don't think cvCornerHarris is even close to what you need.
A much better start would be to experiment with the demo available at: OpenCV-2.3.0/samples/cpp/squares.cpp. This technique uses Canny(), dilate() and findCountour().
Right out of the box, this demo outputs:
I believe that with a few tweaks here and there you can have your party started.