I have a bunch of depth images (see attachement). I want to perform skeletal tracking on them. This data is captured though ASUS Xtion sensor but I only have access to the depth images and not to the videos. Is there a way to perform skeletal tracking in MATLAB/OpenCV or something else on these depth images?
https://qph.cr.quoracdn.net/main-qimg-e538464838f1e0909ec5eda0c13a6956
Hello Mr Khan,
Sir, I think Kinect based tracking may be helpful. You can go through the following papers.
Schwarz, Loren Arthur, et al. "Human skeleton tracking from depth data using geodesic distances and optical flow." Image and Vision Computing 30.3 (2012): 217-226.
Olsen, Martin Aastrup, et al. "Convolution approach for feature detection in topological skeletons obtained from vascular patterns." Computational Intelligence in Biometrics and Identity Management (CIBIM), 2011 IEEE Workshop on. IEEE, 2011.
Regards
Ehtesham
Luckily I had some experience in pattern recognization in depth images. I think the first step is to use the Canny algorithm to detect the edge of people in the picture. When you get the edge image of your depth image, you can just recognize the human body simplely as a circle and a cylinder. This is just a simple way to detect the human body edge, because you can easily find out circles &lines by using Hough transformation. And you also can use other human body's features to improve the performance of body edge detection. Anyway you need to find what part of this image is a person first.Then searching from the head edge, you can estimate the length of the bones and you will get the raw postion of skeleton. If you can get dynamic images, that will absolutely help you get the better position of skeleton.
@Yu Luo I am working on thermal images. suppose, person folds the hands and keeps in front of him. Then what is the way to find the skeleton of hands? thank you
@Vijay Shejal Hi, about the best way, I think the best way of pattern recognition is to teach the computer do what you do. First, you need to figure out that how can a human recognize the hands in the thermal images. What's the features to recognize the hands? Color or Shape? Color , it's just the digital data for the computer. Shape, you can use the gradient of color to get the edges of your thermal images, then you can find a hands-like shape within the edges.
I would like to ask the similar questions. I want to extract skeleton joint positions in video frames.
How can MATLAB save these continuous frames, perform skeleton tracking and returns the skeleton joint position values?
- Badges
- Science topic
- Similar topics
- Methodology
- Sensors