Missing data are common in range images, due to geometric occlusions, limitations in the sensor field of view, poor reflectivity, depth discontinuities, and cast shadows. Using registration to align these data often fails, because points without valid correspondences can be incorrectly matched. This paper presents a maximum likelihood method for registration of scenes with unmatched or missing data. Using ray casting, correspondences are formed between valid and missing points in each view. These correspondences are used to classify points by their visibility properties, including occlusions, field of view, and shadow regions. The likelihood of each point match is then determined using statistical properties of the sensor, such as noise and outlier distributions. Experiments demonstrate a high rates of convergence on complex scenes with varying degrees of overlap.

Investigates the use of Euclidean invariant features in a generalization of iterative closest point (ICP) registration of range images. Pointwise correspondences are chosen as the closest point with respect to a weighted linear combination of positional and feature distances. It is shown that, under ideal noise-free conditions, correspondences formed using this distance function are correct more often than correspondences formed using the positional distance alone. In addition, monotonic convergence to at least a local minimum is shown to hold for this method. When noise is present, a method that automatically sets the optimal relative contribution of features and positions is described. This method trades off the error in feature values due to noise against the error in positions due to misalignment. Experimental results suggest that using invariant features decreases the probability of being trapped in a local minimum and may be an effective solution for difficult range image registration problems where the scene is very small compared to the model.

To construct three dimensional models from range data, the simultaneous registration of multiple views must be accomplished. There exists a natural hierarchy between the local problem of aligning two views and the global problem of maintaining a consistent network of relative orientation and positions between views. Our paper addresses the global problem, where we compute orientation and position estimates from noisy inter-frame measurements. This is accomplished by building a graph of infer-frame relations, selecting a set of basis cycles which fully describe the graph, and distributing accumulated error about the basis cycles.

Investigates the use of Euclidean invariants in the iterative closest point registration of range images. Invariants are used in a modified distance function for the selection of point correspondences. Theoretical results show that under ideal conditions, using invariants can only improve the chance of a making correct correspondences. In addition, monotonic convergence to a local minimum is preserved. Experimental results show that using invariant features accelerates the registration and decreases the probability of being trapped in a local minimum.

The authors present a novel approach to the problem of constructing a depth map from a sequence of monocular images. The approach requires knowledge only of the axial translation component of the moving camera. When the configuration of the camera is known with respect to the mobile platform, this information can be readily obtained by projecting displacement data from a wheel encoder or range sensors onto the focal axis. It is hypothesized that there is a pair of feature points which have the same depth. Based on this hypothesis, the depths are calculated for all pairs of feature points found in the image. As the robot moves, the relative location of two points changes in a specific and predictable manner on the image plane if they are actually at the same depth. The motion of each pair of points is observed, and if it is consistent with the predicted behavior, the hypothesis is accepted. Accepted pairs create a graph structure which contains depth relations among the feature points. The algorithm is robust against rotational and translational motion noises, and its performance has been experimentally demonstrated.>

The infrared (IR) range sensors currently mounted on a mobile pllatform in the Mobile Robot Laboratory at the University of Michigan have a limited range resolution due to their 4 bit analog-to- digital converter (ADC). However, it is possible to obtain better resolution without changing the current hardware configuration by using knowledge about the robot's motion. This paper presents the algorithm for IR range estimation with sub-step accuracy. As the robot moves, a transition monitor detects the instance that a trimmed mean of the IR sensor outputs changes from one ADC step to another and evaluates the importance of the data depending on the type of transition. The information from the transition monitor is integrated recursively in time with the displacement information from an encoder to give the qub-step accuracy in the range estimates. Our algorithm also provides an improved range estimate error bound which is useful in sensor fusion and other applications. Details of the transition monitor are described in this paper, and experimental results which demonstrated a successful implementation of the algorithm are presented. The idea behind the algorithm can be applied to other sensor modalities having resolution primarily limited by a digitization device.