We present a sequential factorization method for recovering the three-dimensional shape of an object and the motion of the camera from a sequence of images, using tracked features. The factorization method originally proposed by Tomasi and Kanade (1992) produces robust and accurate results incorporating the singular value decomposition. However, it is still difficult to apply the method to real-time applications, since it is based on a batch-type operation and the cost of the singular value decomposition is large. We develop the factorization method into a sequential method by regarding the feature positions as a vector time series. The new method produces estimates of shape and motion at each frame. The singular value decomposition is replaced with an updating computation of only three dominant eigenvectors, which can be performed in O(P/sup 2/) time, while the complete singular value decomposition requires O(FP/sup 2/) operations for an F/spl times/P matrix. Also, the method is able to handle infinite sequences, since it does not store any increasingly large matrices. Experiments using synthetic and real images illustrate that the method has nearly the same accuracy and robustness as the original method.

This paper describes the design and the implementation of a high-speed visual tracking system based on local correlation matching. The system consists of only two PC board modules, which conform to VME bus specifications. The video module is used to input and output video signals, while the tracking module is used to perform visual tracking with the local correlation method. The tracking module contains three frame memories, an image flow controller, a DMA controller for image data transfer, and a dedicated LSI to perform the correlation matching of 8-bit gray-scale images. The processing time is 110 microseconds for an 8/spl times/8 pixel template and 290 microseconds for a 16/spl times/16 pixel template. This paper also describes some of our experimental implementations, including human tracking, recognition of hand movements and signs, and traffic monitoring. Compared to conventional general-purpose image processing systems, our hardware system is compact and inexpensive. This high-performance system can be used for a wide variety of motion analysis applications.

The authors developed a small, lightweight rangefinder system which can be mounted on the wrist of a robot. Ranging takes place using a binary-coded projection method based on the principle of triangulation. The head of the range finger comprises a multi-slit projector with an infrared semiconductor laser, a nematic-cholesteric phase-transition (NCPT) liquid crystal shutter, and a small CCD camera. A head size of 170 mm/spl times/80 mm/spl times/50 mm and a weight of 230 grams were achieved. A dedicated image processor was developed to speed up range data generation. The time required to produce a range image is about 0.5 second. The system was tested on a real scene of a mechanical part, which is used in truss assembly in space.

A motion stereo algorithm for measuring straight lines three-dimensionally is presented. The three-dimensional information is obtained by moving a single camera linearly and processing the resulting sequential multiple images. The algorithm utilizes spherical mapping, which is a transformation from a point to a great circle on a sphere. This algorithm makes it possible to process an image from a fisheye lens, which has a wide field of view. A straight line composed of several pixels is concentrated at a single point on a two-dimensional image. A three-dimensional image measurement is then performed by use of the above point. This yields high measurement accuracy and reliability. The measurement accuracy can be improved by lengthening the span of the camera movement and by increasing the number of frames. The measurement reliability can also be improved by increasing the number of frames. The basic algorithm is described, and its characteristics relating to accuracy and reliability are tested.>