nbsrujan's profile - activity

If points which are to be tracked(from first frame), are not present in frames, they are not tracked properly. When there are less points tracked, algorithm present in Opencv examples, initialized itself. Can you post code snippet, so that we can see what's problem is?

Yes, I did try it. Before 2 years.

No need to calibrate cameras only with Chessboards. Instead if you know object coordinates in world coordinate system(can be object) and corresponding image coordinates, it is straightforward to calibrate camera and estimate intrinsic parameters and distortion parameters of camera. You can use calibrateCamera() function. You also get rotation of translation of camera in world's space of all instances.

There is no such function for transforming points from one space to another in OpenCv. Atleast till OpenCv 3.0

There are methods implemented based on learning features, visual cues. You can refer to paper 3-D Reconstruction from a Single Still Image here. Using this method they have created qualitatively correct 3-d models for 64.9% of 588 images downloaded from the internet, and also extended model to produce large scale 3d models from a few images

Many times it was the same to my surprise. But its better to try once.

From the experience, camera id '0' is for default camera. IN your case it will be 'webcam', if it's working fine. If there are several cameras, then id will me related to USB port number.

Instead of using cv2.HOUGH_GRADIENT use cv2.cv.CV_HOUGH_GRADIENT