Detection and matching

When a new estimate of the object's pose is available, the visibility of each edge feature is determined and internal camera parameters are used to project the model of the object onto the image plane. For each visible edge, a number of image points is generated along the edge. So called tracking nodes are assigned at regular intervals in image coordinates along the edge direction. The discretization is performed using the Bresenham algorithm [6]. After that, a search is performed for the maximum discontinuity (nearby edge) in the intensity gradient along the normal direction to the edge. The edge normal is approximated with four directions: -45,0,45,90 degrees. In each point along a visible edge, the perpendicular distance to the nearby edge is determined using a one-dimensional search. The search starts at the projected model point and the traversal continues simultaneously in opposite search directions until the first local maximum is found. After the normal displacements are available, the method proposed in [5] is used. Lie group and Lie algebra formalism are used as the basis for representing the motion of a rigid body and pose estimation. Implementation details can be found in [12]. A few images from a tracking sequence are shown in Fig. .

Figure: first row) An example of tracking a package of raisins: a fairly textured object against a textured background. The estimated pose of the object is overlaid in white. During this experiment a 6mm lens was used and the object was at the distance of approximately 50cm from the camera, and second row) A moving camera and a static object show the ability of the system to cope with significant depth changes and perspective effects.

Figure: A sequence of a 6DOF visual control.