Integration of 21/2D sketches obtained at different observing stations into a consistent world (or object) representation is one of the central issues in computer vision and robotics. The resolution and accuracy of 21/2D sketches may be different from one view point to another, and inconsistent data between different observations may occur. This paper presents one approach to building a 3-D world representation from range image sequences for road scenes including moving objects. First, a range image is transformed into a height map representing the height information from the ground plane on which the observer is assumed to lie, then it is segmented into the ground plane and objects on it. In order to capture the resolution and accuracy of the range data and to reprsent the consistency of the height information between different height maps, we define a reliability of the height information for each grid on the height map. Using the reliability, the system finds the correspondences of both static and moving objects between different observations, and successively refines the height information and its reliability with newly acquired data, dealing with inconsistent data. We show the results applied to road scenes which are physically simulated by landscape toy models using a range finder based on the structured light.

This paper presents a method for spatial and temporal segmentation of long image sequences which include multiple independently moving objects, based on the Minimum Description Length (MDL) principle. By obtaining an optimal motion description, we extract spatiotemporal (ST) segments in the image sequence, each of which consists of edge segments with similar motions. First, we construct a family of 2D motion models, each of which is completely determined by its specified set of equations. Then, based on these sets of equations we formulate the motion description length in a long sequence. The motion state of one object at one moment is determined by finding the model with shortest description length. Temporal segmentation is carried out when the motion state is found to have changed. At the same time, the spatial segmentation is globally optimized in such a way that the motion description of the entire scene reaches a minimum.

A method for detecting multiple rigid motions in images from an optical flow field obtained with multi-scale, multi-orientation filters is proposed. Convolving consecutive gray scale images with a set of eight orientation-selective spatial Gaussian filters yields eight gradient constraint equations for the two components of a flow vector at every location. The flow vector and an uncertainty measure are obtained from these equations. In the neighborhood of motion boundary, the uncertainty of the flow vectors increase. By using multiple sets of filters of different scales, multiple flow vectors are obtained at every location, from which the one with minimal uncertainty measure is selected. The obtained flow field is then segmented in order to solve the aperture problem and to remove noise without blurring discontinuity in the flow field. Discontinuities are first detected as those locations where flow vectors have relatively larger uncertainty measures. Then similar flow vectors are gouped into regions. By modeling flow vectors, regions are merged to form segments each of which belongs to a planar patch of a rigid object in the scene.

This paper describes a method for describing a three-dimensional (3-D) scene by integrating color and range data. Range data is obtained by a feature-based stereo method developed in our laboratory. A color image is segmented into uniform color regions. A plane is fitted to the range data inside a segmented region. Regions are classified into three types based on the range data. A certain types of regions are merged and the others remain unless the region type is modified. The region type is modified if the range data on a plane are selected by removing of the some range data. As a result, the scene is represented by planar surfaces with homogeneous colors. Experimental results for real scenes are shown.

The step size is a parameter of fundamental importance in learning algorithms, particularly for the natural policy gradient (NPG) methods. We derive an upper bound for the step size in an incremental NPG estimation, and propose an adaptive step size to implement the derived upper bound. The proposed adaptive step size guarantees that an updated parameter does not overshoot the target, which is achieved by weighting the learning samples according to their relative importances. We also provide tight upper and lower bounds for the step size, though they are not suitable for the incremental learning. We confirm the usefulness of the proposed step size using the classical benchmarks. To the best of our knowledge, this is the first adaptive step size method for NPG estimation.