Motion parameter estimation by tracking stationary three-dimensional straight lines in image sequences

The purpose of this work is to utilize extracted linear features from imagery collected by a Mobile Mapping System (MMS) for the purpose of motion parameter estimation. The MMS is assumed to be equipped with two or more cameras. The relative relationship between these cameras is rigid, i.e. it does not change from one epoch to the other and can be determined by a calibration procedure. An optimal (minimal) representation of 3-D straight lines that requires only four parameters is discussed. The perspective transformation of the extracted linear features back and forth between image and object space is derived. By tracking stationary linear features in successive image sequences, the motion parameters, rotation and displacement, between the system's position at the corresponding epochs can be estimated. The suggested algorithm works as follows. First, straight-line features are extracted from the imagery. Then, a model space coordinate system is defined for each system position at which an image set is collected. The origin of this system can be chosen as the perspective center of one of the cameras onboard. Then, the linear features in successive image sets are projected into the corresponding model space. It has to be noted that the projection procedure is performed separately for the image sets under consideration. The mathematical model that relates the parameters of the resulting conjugate lines to the motion parameters of the system is derived. The relationship is obtained by imposing four constraints on the four degrees of freedom of any conjugate Lines. The suggested algorithm is used to estimate the motion parameters using conjugate linear features in successive stereo-pairs collected by the GPS van of the Ohio State University. This algorithm will be helpful in deriving the motion parameters of the vehicle during periods of failure of the positioning component onboard, e.g. GPS loss of lock for a long time interval. (C) 1998 Elsevier Science B.V. All rights reserved.