next up previous
Next: Details of the (Edge Up: Tracking Previous: Extraction of Edge Segments

Matching of Data Edge Segments (DESs) to Model Edge Segments (MESs)

There are now two kinds of ESs available for an OC: the DES derived by aggregating EEs obtained from the raw digitized image frame, and ESs from the SS which will be denoted as   Model Edge Segments (MES). The next step to be discussed establishes tentative associations between DES and MES. The Mahalanobis-distances between the parameter vectors for the DESs and the MESs form the basis to compute a correction of the state vector describing the current state estimate for the instantiated OC, see [Koller et al. 93]. The subprocess performing this update-step is indicated by the octagon connected by links between the SS-panel, the DES-panel, and the panel depicting the updated instantiated OC partially occluded by the component panel in Figure 4 related to the initial OC instantiation. The correction of the state-vector x forms part of the update-step of an   Iterated Extended Kalman-Filter (IEKF). The computation of such a correction will be iterated until the remaining sum of residuals between DESs and MESs does not change more than a tolerance.

Since corrections to the OC's state vector x may result in MESs becoming invisible due to (self)occlusion by the opaque polyhedral object model and other - heretofore occluded - ones may become visible in the SS, the association between DESs and MESs has to be re-evaluated following each correction step for the state vector x, see [Koller et al. 93].

Once the IEKF has converged to the final estimate tex2html_wrap_inline1674 of the updated state-vector, assumptions about the motion of the object in the scene can be used to predict the expected state-vector tex2html_wrap_inline1676 at the next (half)frame-time k+1. We formulate our assumptions about vehicle motions in the SD as a   Motion Model (MM): Xtrack assumes that the vehicle moves along a circle. Center point coordinates and radius of this circle have to be estimated during the tracking process, based on the estimates for the reference point position in the road plane, the translational speed and the angular velocity. If the radius estimate exceeds a threshold, infinity will be substituted, implying that the vehicle moves in a pure translation along a straight line.


next up previous
Next: Details of the (Edge Up: Tracking Previous: Extraction of Edge Segments

HHN (Karlsruhe): Conceptual Descriptions from Image Sequences (Dec. 16, 1996)