Frank P. Kuhl

Partial shape recognition using dynamic programming

A partial-shape-recognition technique utilizing local features described by Fourier descriptors is introduced. A dynamic programming formulation for shape matching is developed, and a method for comparison of match quality is discussed. This technique is shown to recognize unknown contours that may be occluded or that may overlap other objects. Precise scale information is not required, and the unknown objects may appear at any orientation with respect to the camera. The segment-matching dynamic programming method is contrasted with other sequence-comparison techniques that utilize dynamic programming. Experimental results are discussed that indicate that partial contours can be recognized with reasonable accuracy. >

A Nonlinear Tracker Using Attitude Measurements

The subject of this paper involves tracking the present position of a maneuvering aircraft as well as predicting its future position. A tracking filter is developed that uses aircraft attitude angles (yaw, pitch, roll) in addition to the usual radar measurements. Computer simulation of tracker performance when tracking violently maneuvering aircraft indicates that a dramatic improvement is obtained by using attitude information. The approach taken is to develop a 12-or 15-state extended Kalman filter that models both translational and rotational degrees of freedom. By measuring and estimating attitude it is possible to approximately determine the magnitude and direction of the force system acting on the vehicle and therefore determine vehicle linear acceleration. Knowledge of acceleration is then used to improve the estimate of present and future position of the vehicle being tracked. Simulation of a T-38 aircraft performing a 5 g turn indicates that the new tracker produces maximum trajectory prediction errors that are 36 percent of the errors experienced by more conventional trackers.

A nonlinear helicopter tracker using attitude measurements

Extended-Kalman-filter-based trackers are discussed for maneuvering helicopters that use body angle and rotor tip-path-plane angle measurements in addition to the usual radar position measurements. Improvements were found in tracker performance when the body rotation and rotor tip-path-plane degrees of freedom were modeled within the extended Kalman filter. Tracker performance was further improved when measurements of body angles and rotor tip-path-plane angles were made available to the tracker. >

Global shape recognition of 3-D objects using a differential library storage

Abstract Shape recognition of three-dimensional rigid objects when viewed by a two-dimensional imaging system is discussed. Fourier descriptor features are used that are invariant to scale, translation, and rotation about the viewing axis. A differential library storage method is developed which allows compact storage of the many shapes generated by viewing an object from different aspect angles. Experimental results are presented using aircraft and normalized Fourier descriptors which show the effectiveness of the differential library storage. An automatic shape learning system is also demonstrated.

Tracking accuracy improvement using noisy target orientation measurements

Abstract This paper discusses to what extent target‐image information can improve the performance of a radar‐based‐target‐tracking system and how accurate this information must be to provide a substantial improvement in tracker performance. The degree of improvement of tracker performance obtained by supplementing radar‐derived position and velocity information with optical‐image‐derived attitude information is investigated for several examples of maneuvering aircraft and ground vehicles. The effects of tracking condition factors such as measurement accuracy, maneuver intensity and prediction time are considered.

Airplane recognition based on silhouettes generated by the stable structure segmentor

Abstract Since 1980, the majority of airplane recognition experiments have employed silhouettes generated from wireframe models of airplanes. From these experiments, the basic silhouette‐based classifiers have matured. A standard classification system is composed of three basic components: a shape library, a feature extractor, and a logic classifier. Additionally, a segmentor component must precede the classifier to obtain a silhouette from a real world image. The features extracted from the silhouette are normalized to remove variations with respect to scale, rotation, and translation. They are then matched against the elements in the shape library to obtain recognition. The experiments reported in this paper differ from the conventional ones in two respects. Firstly, a segmentor is employed to generate observed silhouettes. Secondly, video images, rather than wireframe models, are used in the analysis. This paper describes a standard recognition system, a segmentation algorithm developed by Hsu (see Hsu...

Radar recognition of geometric classes in free space

Abstract A method for recognizing targets at any aspect angle is proposed using polarized, radar‐backscatter measurements. The measurements are taken at wavelengths commensurate with significant target‐body dimensions; i.e., near body resonance. The method applies to targets of arbitrary shape that cannot deform, such as missiles. The objects can be of arbitrary size and refractive index. The method is easily extended to the recognition of classes of objects that differ only by a scale factor in size. This is accomplished by scanning the radar frequency and making feature comparisons against stored libraries of features taken at several wavelengths. These torus‐like‐surface feature libraries are described by use of an n‐dimensional chain code, and a quantization interval for the classification space is specified according to the spatial frequency spectra of these toroids.

Spherical harmonies and moments for recognition of three‐dimensional objects

Abstract Many two‐dimensional shape recognition techniques have appeared in the literature for identification of objects from intensity images. In this paper, two of these techniques will be extended to the recognition of objects from three‐dimensional (3‐D) data. Specifically, the concepts of Fourier Descriptors will be extended to three dimensions through the use of spherical harmonics. The use of moments will also be extended to accommodate 3‐D data. Feature extraction, normalization, and classification will be discussed for both techniques. Recognition experiments and results will be presented. The experimental results indicate that the use of 3‐D data improves classification accuracy.

Shape analysis using generic features

Abstract Generic shape features are developed and tested that are based on Fourier descriptors of the boundary curve. The harmonic components of the boundary contour provide interesting information about generic shape features. Those implemented include: similarity to a rectangle, square, circle, and ellipse; bilateral symmetry, axial symmetry, and N‐fold symmetry. Confidence numbers ranging from 0 to 100 are calculated for each generic feature so that the confidence numbers have correlation with a humans perception of the presence of each generic feature in the overall shape. Generic shape attributes can be assigned without any prior training or library storage which is required for other kinds of features, so they are very useful in automatic recognition of objects based only on general shape properties. A rule‐based system for the use of generic shapes in shape classification has been implemented. Weighted combinations of generic features can be specified by user‐defined rules, which can include a ran...