3-D motion estimation using range data

Advanced vehicle-based safety and warning systems use laser scanners to measure road geometry (position and curvature) and range to obstacles in order to warn a driver of an impending crash and/or to activate safety devices (air bags, brakes, and steering). In order to objectively quantify the performance of such a system, the reference system must be an order of magnitude more accurate than the sensors used by the warning system. This can be achieved by using high-resolution range images that can accurately perform object tracking and velocity estimation. Currently, this is very difficult to achieve when the measurements are taken from fast moving vehicles. Thus, the main objective is to improve motion estimation, which involves both the rotational and translation movements of objects. In this respect, an innovative recursive motion-estimation technique that can take advantage of the in-depth resolution (range) to perform accurate estimation of objects that have undergone three-dimensional (3-D) translational and rotational movements is presented. This approach iteratively aims at minimizing the error between the object in the current frame and its compensated object using estimated-motion displacement from the previous range measurements. In addition, in order to use the range data on the nonrectangular grid in the Cartesian coordinate, two approaches have been considered: 1) membrane fit, which interpolates the nonrectangular grid to the rectangular grid, and 2) the nonrectangular-grid range data by employing derivative filters and the proposed transformation between the Cartesian coordinates and the sensor-centered coordinates. The effectiveness of the proposed scheme is demonstrated for sequences of moving-range images.