You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I will make the point of the WebARKit library. Actually it is in the dev branch of this repository, before merging in master and to be officially released it is required to add some important features.
One of the most important feature to add is a better pose matrix solution. We receive the pose matrix from the knn matcher but it needs to be evaluated with a PnP solver. This require a camera matrix for the PnP pose computation, that for now our solution is lacking, and after that use a Kalman filter (or other filter) to stabilize the 6DOF tracking result.
Adding to this: in order to utilize the pose matrix we need a set of functions for GL graphics, to convert the pose matrix as Right hand GL matrix and also a function to compute the Camera Projection matrix.
Summing up, we need to develop:
A new WebARKitCamera class was created in PR #22, before procede to the second task, i considered that it is required to create a class that handle the "pattern", the image that is tracked. This class should contain the Point2d and Point3d data and of course keypoints and descriptors as well. this was implemented in OCVTartoolkitx but i saw also in another code projects.
I merged the PnP related code with #23 but it think there is something wrong with buildOpticalFlowPyramid and calcOpticalFlowPyrLK. Probably the steps are not in the right sequence and this may produce weird results, in computing homography and pose matrix.
I will make the point of the WebARKit library. Actually it is in the
dev
branch of this repository, before merging in master and to be officially released it is required to add some important features.One of the most important feature to add is a better pose matrix solution. We receive the pose matrix from the
knn
matcher but it needs to be evaluated with aPnP
solver. This require a camera matrix for the PnP pose computation, that for now our solution is lacking, and after that use a Kalman filter (or other filter) to stabilize the 6DOF tracking result.Adding to this: in order to utilize the pose matrix we need a set of functions for GL graphics, to convert the pose matrix as Right hand GL matrix and also a function to compute the Camera Projection matrix.
Summing up, we need to develop:
arGLposeMatrix
andcameraProjectionMatrix
calcOpticalFlowPyrLK
approach -> done in PR Improving c++ code #27The text was updated successfully, but these errors were encountered: