I am using a web cam to get video feed and then performing motion tracking on this video feed. The motion tracker returns (x,y) co-ordinates continuously.
I want to use these (x,y) to recognize gestures such as "swipe left", "swipe right", "swipe up" or "swipe down".
How do i make and store templates of these gestures and how do i figure out/recognize if one
of the gestures has happened ?
Thank you in advance :)
PS: I am using Flex 4 and ActionScript 3.0. If someone could help me out with the logic, i can write it in ActionScript.
An approach I could think of working would be to have a series of (X,Y) coordinates representing points along the gesture. On a small scale if a gesture that passed through your screen was graphed as such:
|1|1|1|
|1|0|0|
|1|0|0|
and represented (from the upper left corner representing 0,0):
(0,2)(0,1)(0,0)(0,1)(0,2)
Break the x,y coordinates up into individual 2d arrays with total distance traveled between the current coordinate and the first point (in all cases in this example it would increment by 1) so you would have two arrays:
X:(0,0)(1,1)(2,2)
Y:(0,1)(1,1)(2,2)
Now do a least square fit on each array to find the closest representation of the change in x and change in y as quadratic functions. Do the same to your per-deteremined gestures and then plug in the x,y coordinates of your per-determined gestures into to the user's gesture's quadratic function and the per-determined gestures you designed and see which one it matches the closest. This is your gesture.
(I've never tried processing gestures, but I don't see why this wouldn't work)
You should divide your task into smaller subtasks. In computer vision there is no thing like a generic gesture detection that works out of the box in all environments.
First of all, you need to be able to detect motion at all. There are several ways to do this, e.g. background subtraction or blob tracking.
Then you need to extract certain features from your image, e.g. a hand. Again, there is more than one way to do this. Starting from skin color approximation/evaluation, which is very error prone to different lighting conditions, to more advanced techniques which really try to "analyze" the shape of an object. Those algorithms "learn" over time how a hand should look like.
I can only recommend to buy a decent book about computer vision and to research the web for articles etc. There are also libraries like OpenCV you can use for learning more about the implementation side. There should be several ports of OpenCV to ActionScript 3. I also can recommend the articles and tools from Eugene Zatepyakin (http://blog.inspirit.ru). He's doing great CV stuff with ActionScript 3.
Long story short, you should research motion tracking and feature extraction.
The best place to start is to read about how sign language recognition or trackpad input works, such as creating reference images and comparing them to user input. Specific to Adobe, there's the FLARToolKit, which is detailed in an augmented reality article on their website.
References:
Trackpad Science
Hand Gesture Recognition
Sign Language Recognition Research - PDF
Gesture Recognition Walkthrough - Video
Related
I have an android application which is getting gesture coordinates (3 axis - x,y,z) from an external source. I need to compare them with coordinates which I have in my DB and determine whether gesture is the same/similar or not.
I also need to add some tolerance, since accelerometer (device which captures gestures) is very sensitive. It would be easy, but I also want to consider e.g. "big circle" drawn in the air, same as "small circle" drawn in the air. meaning that there would be different values, but structure of the graph would be the same, right?
I did some research and I found out I could use Dynamic time wraping possibly in combination with Hidden Markov model.
I tried to do some further reading on it, but I didn't find much. I just found a Java library called FastDTW but can't figure out which methods are the right ones for my problem.
Would you please provide me any lead on this?
See these images, they are the same gesture but one is done in a slower motion than another.
Faster one:
Slower one:
I haven't captured images of the same gesture where one would be smaller than another, might add that later.
I want to remove background and get deer as a foreground image.
This is my source image captured by trail camera:
This is what I want to get. This output image can be a binary image or RGB.
I worked on it and try many methods to get solution but every time it failed at specific point. So please first understand what is my exact problem.
Image are captured by a trail camera and camera is motion detector. when deer come in front of camera it capture image.
Scene mode change with respect to weather changing or day and night etc. So I can't use frame difference or some thing like this.
Segmentation may be not work correctly because Foreground (deer) and Background have same color in many cases.
If anyone still have any ambiguity in my question then please first ask me to clear and then answer, it will be appreciated.
Thanks in advance.
Here's what I would do:
As was commented to your question, you can detect the dear and then perform grabcut to segment it from the picture.
To detect the dear, I would couple a classifier with a sliding window approach. That would mean that you'll have a classifier that given a patch (can be a large patch) in the image, output's a score of how much that patch is similar to a dear. The sliding window approach means that you loop on the window size and then loop on the window location. For each position of the window in the image, you should apply the classifier on that window and get a score of how much that window "looks like" a dear. Once you've done that, threshold all the scores to get the "best windows", i.e. the windows that are most similar to a dear. The rational behind this is that if we a dear is present at some location in the image, the classifier will output a high score at all windows that are close/overlap with the actual dear location. We would like to merge all that locations to a single location. That can be done by applying the functions groupRectangles from OpenCV:
http://docs.opencv.org/modules/objdetect/doc/cascade_classification.html#grouprectangles
Take a look at some face detection example from OpenCV, it basically does the same (sliding window + classifier) where the classifier is a Haar cascade.
Now, I didn't mention what that "dear classifier" can be. You can use HOG+SVM (which are both included in OpenCV) or use a much powerful approach of running a deep convulutional neural network (deep CNN). Luckily, you don't need to train a deep CNN. You can use the following packages with their "off the shelf" ImageNet networks (which are very powerful and might even be able to identify a dear without further training):
Decaf- which can be used only for research purposes:
https://github.com/UCB-ICSI-Vision-Group/decaf-release/
Or Caffe - which is BSD licensed:
http://caffe.berkeleyvision.org/
There are other packages of which you can read about here:
http://deeplearning.net/software_links/
The most common ones are Theano, Cuda ConvNet's and OverFeat (but that's really opinion based, you should chose the best package from the list that I linked to).
The "off the shelf" ImageNet network were trained on roughly 10M images from 1000 categories. If those categories contain "dear", that you can just use them as is. If not, you can use them to extract features (as a 4096 dimensional vector in the case of Decaf) and train a classifier on positive and negative images to build a "dear classifier".
Now, once you detected the dear, meaning you have a bounding box around it, you can apply grabcut:
http://docs.opencv.org/trunk/doc/py_tutorials/py_imgproc/py_grabcut/py_grabcut.html
You'll need an initial scribble on the dear to perform grabcu. You can just take a horizontal line in the middle of the bounding box and hope that it will be on the dear's torso. More elaborate approaches would be to find the symmetry axis of the dear and use that as a scribble, but you would have to google, research an implement some method to extract symmetry axis from the image.
That's about it. Not straightforward, but so is the problem.
Please let me know if you have any questions.
Try OpenCV Background Substraction with Mixture of Gaussians models. They should be adaptable enough for your scenes. Of course, the final performance will depend on the scenario, but it is worth trying.
Since you just want to separate the background from the foreground I think you do not need to recognize the deer. You need to recognize an object in motion in the scene. You just need to separate what is static in a significant interval of time (background) from what is not static: the deer.
There are algorithms that combine multiple frames from the same scene in order to determine the background, like THIS ONE.
You mentioned that the scene mode changes with respect to weather changing or day and night considering photos of different deers.
You could implement a solution when motion is detected, instead of taking a single photo, it could take a few ones with some interval of time.
This interval has to be long as to get the deer in different positions or out of the scene and at the same time short enough to not be much affected by scene variations. Perhaps you need to deal with some brightness variation, but I think it is feasible to determine the background using these frames and finally segment the deer in the "motion frame".
Is it possible to make a 3D representation of an object by capturing many different angles using a webcam? If it is, how is it possible and how is the image-processing done?
My plan is to make a 3D representation of a person using a webcam, then from the 3D representation, i will be able to tell the person's vital statistics.
As Bart said (but did not post as an actual answer) this is entirely possible.
The research topic you are interested in is often called multi view stereo or something similar.
The basic idea resolves around using point correspondences between two (or more) images and then try to find the best matching camera positions. When the positions are found you can use stereo algorithms to back project the image points into a 3D coordinate system and form a point cloud.
From that point cloud you can then further process it to get the measurements you are looking for.
If you are completely new to the subject you have some fascinating reading to look forward to!
Bart proposed Multiple view geometry by Hartley and Zisserman, which is a very nice book indeed.
As Bart and Kigurai pointed out, this process has been studied under the title of "stereo" or "multi-view stereo" techniques. To be able to get a 3D model from a set of pictures, you need to do the following:
a) You need to know the "internal" parameters of a camera. This includes the focal length of the camera, the principal point of the image and account for radial distortion in the image.
b) You also need to know the position and orientation of each camera with respect to each other or a "world" co-ordinate system. This is called the "pose" of the camera.
There are algorithms to perform (a) and (b) which are described in Hartley and Zisserman's "Multiple View Geometry" book. Alternatively, you can use Noah Snavely's "Bundler" http://phototour.cs.washington.edu/bundler/ software to also do the same thing in a very robust manner.
Once you have the camera parameters, you essentially know how a 3D point (X,Y,Z) in the world maps to an image co-ordinate (u,v) on the photo. You also know how to map an image co-ordinate to the world. You can create a dense point cloud by searching for a match for each pixel on one photo in a photo taken from a different view-point. This requires a two-dimensional search. You can simplify this procedure by making the search 1-dimensional. This is called "rectification". You essentially take two photos and transform then so that their rows correspond to the same line in the world (simplified statement). Now you only have to search along image rows.
An algorithm for this can be also found in Hartley and Zisserman.
Finally, you need to do the matching based on some measure. There is a lot of literature out there on "stereo matching". Another word used is "disparity estimation". This is basically searching for the match of pixel (u,v) on one photo to its match (u, v') on the other photo. Once you have the match, the difference between them can be used to map back to a 3D point.
You can use Yasutaka Furukawa's "CMVS" or "PMVS2" software to do this. Or if you want to experiment by yourself, openCV is a open-source computer vision toolbox to do many of the sub-tasks required for this.
This can be done with two webcams in the same ways your eyes work. It is called stereoscopic vision.
Have a look at this:
http://opencv.willowgarage.com/documentation/camera_calibration_and_3d_reconstruction.html
An affordable alternative to get 3D data would be the Kinect camera system.
Maybe not the answer you are hoping for but Microsoft's Kinect is doing that exact thing, there are some open source drivers out there that allow you to connect it to your windows/linux box.
I have a web cam that takes a picture every N seconds. This gives me a collection of images of the same scene over time. I want to process that collection of images as they are created to identify events like someone entering into the frame, or something else large happening. I will be comparing images that are adjacent in time and fixed in space - the same scene at different moments of time.
I want a reasonably sophisticated approach. For example, naive approaches fail for outdoor applications. If you count the number of pixels that change, for example, or the percentage of the picture that has a different color or grayscale value, that will give false positive reports every time the sun goes behind a cloud or the wind shakes a tree.
I want to be able to positively detect a truck parking in the scene, for example, while ignoring lighting changes from sun/cloud transitions, etc.
I've done a number of searches, and found a few survey papers (Radke et al, for example) but nothing that actually gives algorithms that I can put into a program I can write.
Use color spectroanalisys, without luminance: when the Sun goes down for a while, you will get similar result, colors does not change (too much).
Don't go for big changes, but quick changes. If the luminance of the image changes -10% during 10 min, it means the usual evening effect. But when the change is -5%, 0, +5% within seconds, its a quick change.
Don't forget to adjust the reference values.
Split the image to smaller regions. Then, when all the regions change same way, you know, it's a global change, like an eclypse or what, but if only one region's parameters are changing, then something happens there.
Use masks to create smart regions. If you're watching a street, filter out the sky, the trees (blown by wind), etc. You may set up different trigger values for different regions. The regions should overlap.
A special case of the region is the line. A line (a narrow region) contains less and more homogeneous pixels than a flat area. Mark, say, a green fence, it's easy to detect wheter someone crosses it, it makes bigger change in the line than in a flat area.
If you can, change the IRL world. Repaint the fence to a strange color to create a color spectrum, which can be identified easier. Paint tags to the floor and wall, which can be OCRed by the program, so you can detect wheter something hides it.
I believe you are looking for Template Matching
Also i would suggest you to look on to Open CV
We had to contend with many of these issues in our interactive installations. It's tough to not get false positives without being able to control some of your environment (sounds like you will have some degree of control). In the end we looked at combining some techniques and we created an open piece of software named OpenTSPS (Open Toolkit for Sensing People in Spaces - http://www.opentsps.com). You can look at the C++ source in github (https://github.com/labatrockwell/openTSPS/).
We use ‘progressive background relearn’ to adjust to the changing background over time. Progressive relearning is particularly useful in variable lighting conditions – e.g. if lighting in a space changes from day to night. This in combination with blob detection works pretty well and the only way we have found to improve is to use 3D cameras like the kinect which cast out IR and measure it.
There are other algorithms that might be relevant, like SURF (http://achuwilson.wordpress.com/2011/08/05/object-detection-using-surf-in-opencv-part-1/ and http://en.wikipedia.org/wiki/SURF) but I don't think it will help in your situation unless you know exactly the type of thing you are looking for in the image.
Sounds like a fun project. Best of luck.
The problem you are trying to solve is very interesting indeed!
I think that you would need to attack it in parts:
As you already pointed out, a sudden change in illumination can be problematic. This is an indicator that you probably need to achieve some sort of illumination-invariant representation of the images you are trying to analyze.
There are plenty of techniques lying around, one I have found very useful for illumination invariance (applied to face recognition) is DoG filtering (Difference of Gaussians)
The idea is that you first convert the image to gray-scale. Then you generate two blurred versions of this image by applying a gaussian filter, one a little bit more blurry than the first one. (you could use a 1.0 sigma and a 2.0 sigma in a gaussian filter respectively) Then you subtract from the less-blury image, the pixel intensities of the more-blurry image. This operation enhances edges and produces a similar image regardless of strong illumination intensity variations. These steps can be very easily performed using OpenCV (as others have stated). This technique has been applied and documented here.
This paper adds an extra step involving contrast equalization, In my experience this is only needed if you want to obtain "visible" images from the DoG operation (pixel values tend to be very low after the DoG filter and are veiwed as black rectangles onscreen), and performing a histogram equalization is an acceptable substitution if you want to be able to see the effect of the DoG filter.
Once you have illumination-invariant images you could focus on the detection part. If your problem can afford having a static camera that can be trained for a certain amount of time, then you could use a strategy similar to alarm motion detectors. Most of them work with an average thermal image - basically they record the average temperature of the "pixels" of a room view, and trigger an alarm when the heat signature varies greatly from one "frame" to the next. Here you wouldn't be working with temperatures, but with average, light-normalized pixel values. This would allow you to build up with time which areas of the image tend to have movement (e.g. the leaves of a tree in a windy environment), and which areas are fairly stable in the image. Then you could trigger an alarm when a large number of pixles already flagged as stable have a strong variation from one frame to the next one.
If you can't afford training your camera view, then I would suggest you take a look at the TLD tracker of Zdenek Kalal. His research is focused on object tracking with a single frame as training. You could probably use the semistatic view of the camera (with no foreign objects present) as a starting point for the tracker and flag a detection when the TLD tracker (a grid of points where local motion flow is estimated using the Lucas-Kanade algorithm) fails to track a large amount of gridpoints from one frame to the next. This scenario would probably allow even a panning camera to work as the algorithm is very resilient to motion disturbances.
Hope this pointers are of some help. Good Luck and enjoy the journey! =D
Use one of the standard measures like Mean Squared Error, for eg. to find out the difference between two consecutive images. If the MSE is beyond a certain threshold, you know that there is some motion.
Also read about Motion Estimation.
if you know that the image will remain reletivly static I would reccomend:
1) look into neural networks. you can use them to learn what defines someone within the image or what is a non-something in the image.
2) look into motion detection algorithms, they are used all over the place.
3) is you camera capable of thermal imaging? if so it may be worthwile to look for hotspots in the images. There may be existing algorithms to turn your webcam into a thermal imager.
In a multi-touch environment, how does gesture recognition work? What mathematical methods or algorithms are utilized to recognize or reject data for possible gestures?
I've created some retro-reflective gloves and an IR LED array, coupled with a Wii remote. The Wii remote does internal blob detection and tracks 4 points of IR light and transmits this information to my computer via a bluetooth dongle.
This is based off Johnny Chung Lee's Wii Research. My precise setup is exactly like the graduate students from the Netherlands displayed here. I can easily track 4 point's positions in 2d space and I've written my basic software to receive and visualize these points.
The Netherlands students have gotten a lot of functionality out of their basic pinch-click recognition. I'd like to take it a step further if I could, and implement some other gestures.
How is gesture recognition usually implemented? Beyond anything trivial, how could I write software to recognize and identify a variety of gestures: various swipes, circular movements, letter tracing, etc.
Gesture recognition, as I've seen it anyway, is usually implemented using machine learning techniques similar to image recognition software. Here's a cool project on codeproject about doing mouse gesture recognition in c#. I'm sure the concepts are quite similar since you can likely reduce the problem down to 2D space. If you get something working with this, I'd love to see it. Great project idea!
One way to look at it is as a compression / recognition problem. Basically, you want to take a whole bunch of data, throw out most of it, and categorize the remainder. If I were doing this (from scratch) I'd probably proceed as follows:
work with a rolling history window
take the center of gravity of the four points in the start frame, save it, and subtract it out of all the positions in all frames.
factor each frame into two components: the shape of the constellation and the movement of it's CofG relative to the last frame's.
save the absolute CofG for the last frame too
the series of CofG changes gives you swipes, waves, etc.
the series of constellation morphing gives you pinches, etc.
After seeing your photo (two points on each hand, not four points on one, doh!) I'd modify the above as follows:
Do the CofG calculation on pairs, with the caveats that:
If there are four points visible, pairs are chosen to minimize the product of the intrapair distances
If there are three points visible, the closest two are one pair, the other one is the other
Use prior / following frames to override when needed
Instead of a constellation, you've got a nested structure of distance / orientation pairs (i.e., one D/O between the hands, and one more for each hand).
Pass the full reduced data to recognizers for each gesture, and let them sort out what they care about.
If you want to get cute, do a little DSL to recognize the patterns, and write things like:
fire when
in frame.final: rectangle(points)
and
over frames.final(5): points.all (p => p.jerk)
or
fire when
over frames.final(3): hands.all (h => h.click)
A video of what has been done with this sort of technology, if anyone is interested?
Pattie Maes demos the Sixth Sense - TED 2009
Most simple gesture-recognition tools I've looked at use a vector-based template to recognize them. For example, you can define right-swipe as "0", a checkmark as "-45, 45, 45", a clockwise circle as "0, -45, -90, -135, 180, 135, 90, 45, 0", and so on.
Err.. I've been working on gesture recognition for the past year or so now, but I don't want to say too much because I'm trying to patent my technology :) But... we've had some luck with adaptive boosting, although what you're doing looks fundamentally different. You only have 4 points of data to process, so I don't think you really need to "reduce" anything.
What I would investigate is how programs like Flash turn a freehand drawn circle into an actual circle. It seems like you could track the points for duration of about a second, and then "smooth" the path in some fashion, and then you could probably get away with hardcoding your gestures (if you make them simple enough). Otherwise, yes, you're going to want to use a learning algorithm. Neural nets might work... I don't know. Just tossing out ideas :) Maybe look at how OCR is done too... or even Hough transforms. It looks to me like this is a problem of recognizing shapes more than it is of recognizing gestures.
I'm not very well versed in this type of mathematics, but I have read somewhere that people sometimes use Markov Chains or Hidden Markov Models to do Gesture Recognition.
Perhaps someone with a little more background in this side of Computer Science can illuminate it further and provide some more details.