Machine perception of dynamic scenes becomes more and more important for autonomous vehicles and vision-based driver-assistance systems. Even with other 3D ranging devices, dense, detailed and instantaneous detection of optical flow is essential for early distinguishing small moving objects in the 3D environment from the moving vehicle. To overcome the limited performance in the immediacy, resolution, accuracy and acuity of existing methods, we provide an optical flow detection scheme based on a three-phase correlation image sensor (3PCIS) that is capable of Fourier-coefficient imaging combined with an exact and direct algorithm derived from the weighted integral method of identifying the differential equation model from a short-duration observation. To utilize inherent performances of the detection scheme by removing the large and rapid disturbances induced by the rotational fluctuations of the platform, we introduce a software operation of gaze in which the image coordinates are fixed on and smoothly pursue a forward stable object so that the optical flow field is relative to the moving coordinate system. In it, the gaze subsystem continuously provides angular velocity and pose between the camera and gaze target, while the imaging subsystem instantaneously obtains two optical flow distributions by cancelling the ego-rotation components and then removing the outwardly diverging components derived mainly from stationary 3D environments. Possible anomalies captured in each frame instantaneously provide candidates of hazardous objects that should be tracked and further investigated. We examine the performance of optical flow stabilization and anomaly detection using image sequences of monocular 3PCIS mounted on a moving vehicle on town roads and a highway.

Radio frequency (RF) tags are growing popular for many industrial applications such as product tracking and security. We investigate the benefit of RF tagging in an application of 3D object registration from single view. We present a machine vision system that uses a RF Tag device to detect objects prior to locating them visually. The tag system consists of a tag reader that can interrogate, and receive radio signals from, tags attached on objects and characterizing them. Laying the basis of an object model database shared on a network, we perform a knowledge-based recognition task where the information retrieved from the database query serves as the prior. The recognition algorithm used is a matching with projective invariants. Our architecture is a prototype that prefigures tag based applications where physical and logic representations of objects are combined.

Visual tags, of which barcodes represent a successful effort for standardization, have often been used to deepen the level of automation in handling industrial objects. In the past few years, a new type of tags has been the object of intensive developments: radio-frequency (RF) Tags. We present the architecture of a machine vision system that uses RF tag identification (RFID). RFID enables detecting the presence of an object, which polyhedral model and properties are remotely retrieved from a knowledge network database. This model is used for 3D scene analysis by projective geometry. Our system not only facilitates the problem of object identification and registration to a model, but also allows new tag-based applications to be built around the concept of ubiquitous objects.

The authors first analyze dynamic properties of rapidly adapting mechanoreceptors in order to derive a principle of tactile displays. It is shown that Meissner corpuscles with coiled axons and Pacinian corpuscles with layered lamellae are suited to detect equivoluminal distortion of skin. Combining these analyses with a model of a contact and a relative motion between an object and the skin, it is shown that a prerequisite for tactile displays is to generate sources of shear stress that can be spatially dispersed at the skin surface and temporally modulated with a stick-slip frequency determined by parameters of the object to be displayed. As a device that satisfies this prerequisite, we propose a tactile display using Surface Acoustic Waves (SAW). The roughness of the surface can be changed continuously by controlling the burst frequency of the SAW.

Edges, corners, and vertices in an image correspond to 1D (one-dimensional) and 2D discontinuities in the intensity surface of the underlying scene. Ridges and peaks correspond to 1D and 2D extrema in it. All of them can be characterized by the distribution of gradients, particularly by the dimensionality of it. The approach to image field categorization here is to construct a covariance matrix of the gradient vector in each small window and apply the canonical correlation analysis to it. Schwarz's inequality on the matrix determinant and the related differential equation is the key to this analysis. We obtain two operators P/sub EG/ and Q/sub EG/ to categorize the image field into a unidirectionally varying region (UNIVAR), an omidirectionally varying region (OMNIVAR), and a nonvarying region. We investigate the conditions under which their absolute maximum response, i.e. P/sub EG/=1 and Q/sub EG/=1, occurs in the small window and show that they are, respectively, the desired 1D and 2D discontinuities/extrema and OMNIVAR, is in many cases, a 1D pattern in polar coordinates. This leads to an algorithm to obtain further classification and accurate localization of them into edges, ridges, peaks, corners, and vertices through detailed analysis in the informative (varying) axis of them. We examined and compared the performance of the operators and the localization algorithm on various types of images and various noise levels. The results indicate that the proposed method is superior with respect to stability, localization, and resolution.

We propose optimal gradient operators based on a newly derived consistency criterion. This criterion is based on an orthogonal decomposition of the difference between a continuous gradient and discrete gradients into the intrinsic smoothing effect and the self-inconsistency involved in the operator. We show that consistency assures the exactness of gradient direction of a locally 1D pattern in spite of its orientation, spectral composition, and sub-pixel translation. Stressing that inconsistency reduction is of primary importance, we derive an iterative algorithm which leads to accurate gradient operators of arbitrary size. We compute the optimum 3/spl times/3, 4/spl times/4, and 5/spl times/5 operators, compare them with conventional operators and examine the performance for one synthetic and several real images. The results indicate that the proposed operators are superior with respect to accuracy, bandwidth and isotropy.

The most important features of quasi-harmonic signals such as speech or musical sounds are the instantaneously varying amplitude (AM) and pitch (FM) of the harmonics. However, when we want to recognise them after subband decomposition like a human cochlea, the problem is that AM-FM components of the original harmonics are disturbed both by the filtering and the interference between harmonics. In this paper, we propose a method of logarithmic differential decomposition (LDD) and a new AM-FM extraction algorithm based of it. It extracts separately the conventional AM-FM which are common to all the harmonics and the interferometric AM-FM which are synchronous with pitch, hence can be an efficient clue to determine the fundamental. We investigate the role of those features by experiments for speech.

This paper describes a new method for computational auditory scene analysis which is based on (1) waveform operators to extract instantaneous frequency (IF), frequency change (FM), and amplitude change (AM) from subband signals, and (2) the introduction of a voting method into a probability distribution function to extract coherency (shared fundamental frequency, shared FM, and shared AM) involved in them. We introduce non-parametric Kalman filtering for the time-axis integration. A consistent AM operator which is independent of frequency change is newly defined. The sharpness of the resultant probability distribution is examined with relation to the definition of the operators and subband bandwidth. We evaluate the performance of the algorithm by using several speech sounds.

In this paper we propose a new tactile sensing element, acoustic resonant tensor cell. The structure has only a spherical cavity in an elastic tactile sensor body with two fine ultrasound paths extended respectively to an ultrasound transmitter and receiver which are placed at the bottom of the elastomer. The acoustic resonant frequency of the air in the cavity has an explicit relation with the principal stresses around it, which is easily detected by the ultrasound transducers. The sensor is simple and elastic. The single structure presents useful tactile features of multidimensions. The principle and experimental results are described.

Abe and Ando (see Proc. IEEE ICASSP95, p.1368-71, 1995) proposed nonlinear operators which decompose a changing energy of sound in the wavelet domain into three orthogonal components: i.e., loudness and pitch as coherent changes, and timbre as an incoherent change. They showed that they could detect the discontinuity of a single sound stream with excellent temporal resolution and sensitivity. In this paper, they extend the coherency principle so that it can describe and pursue the individual coherency of non-overlapping sound streams in the wavelet domain. It is realized by Parzen's non-parametric estimates and Kalman filtering of the loudness change rate and the pitch shift rate. Using this method, they show some experiments for the extraction of the most salient stream from multiple sound streams.

We propose a new harmonics tracking algorithm for auditory scene analysis. It extracts the most significant stream from multiple sound streams as well as the tracking of each harmonic components. For the optimum tracking of pitch candidates, a novel estimation technique based on nonlinear Kalman filtering and Parzen estimate (non-parametric Kalman filter, NPKF) is used. For invariant reconstruction of harmonic components against pitch estimation error, we perform a filtering process in the wavelet domain. Using simulated data and real world data, we show several experimental results for extracting the most dominant sound stream among multiple ones.

This paper describes a new tactile sensor using acoustic cell structure based on the detection of ultrasound wavefronts in a flexible medium. The acoustic cell consists of a 2/spl times/2 ultrasonic transmitter matrix and a 2/spl times/2 ultrasonic sensing matrix which are placed face to face with an interval of a few tens of wavelength. These cells are embedded in flexible spherical finger-tip-like body of a tactile sensor. First, we show an ultrasonic transduction principle to discriminate small deformation of the cell in five dimensions, i.e. 3D displacement and a pair of surface angles of the transmitter matrix relative to the receiver. We then describe a tactile sensor using a single element of the cell, and show that the sensor can measure: 1) 10 /spl mu/m displacement by 18.5 mm cell height, and 2) 0.001 rad change of surface inclination. We apply the sensor to a robot finger which can sense not only its contact force with arbitrary orientation but also predict the slip of the grasped object.

We propose a method for decomposing instantaneous changes of sounds into three energy components, i.e., loudness, pitch, and timbre. These operators are derived from an eigenstructure analysis of the time-frequency gradient space (a 3-D space spanned by a modulus and partial derivatives of a wavelet transform). By several experiments, we found that they have superior resolution and sensitivity for segmenting speech into phonemes, characterizing dynamical nature of musical sounds, and so on.

Proposes a new tactile sensory structure using integrated sensing elements which the authors call a tensor cell. It is located sparsely in a flexible body of a tactile sensor, and samples there the complete information of a stress tensor which requires six degrees of freedom to describe it. The authors clarify first the advantage of this architecture theoretically based on elasticity theory and matrix algebra. Then the design of the actual tensor cell and experimental results are described, and an application of it to contact surface characterization is shown.

The blind superresolution algorithm performs simultaneously the estimating task of a blur point spread function (PSF) and extrapolating task of the the graded resolution. In this paper, we apply this algorithm to an image recovery problem by using salient edge location information as a convex condition. On the first step, we extract blur-invariant image features, which actually are the steepest lines and points of isolated edges and corners. On the second step, we make a convex set from the blur-invariant image features and obtain a projection operator correspondent to it. By using this with other a prior information, we estimate and recover simultaneously the PSF and the original image through the well-known mathematical projection technique (POCS). Several simulation results are shown in comparison with the other method.>

This paper describes our primary study for a new type of tactile sensing devices which are based on instantaneous localization of acoustic emissions caused necessarily by touch and/or contact movement. We assume the sensor structure which consists of a flexible spherical fingertip-like body and a quadruple PVDF sound sensing matrix is placed at a center of the body. The PVDF matrix works as a wideband acoustic emission (AE) transducers which is capable to detect the arrival direction of a wave packet. Therefore it can resolve and localize multiple ultrasonic emissions at a sensor surface caused by a touch and slip. In order to realize this type of tactile sensor, we present the theory of wave generation at collision, and apply the newly developed sound localization algorithm for an auditory sensor. By assuming the use of this algorithm, and from several experiments under various contact and slip conditions, we show the following features and tactile information: 1) quick localization of touch and detouch, 2) sensation of texture under movement, and 3) quick detection of precursor of slip.>

A new tactile sensing algorithm based on the elastic transfer function of surface deformation is proposed. Utilizing the inherent mechanism of an elastic body, it is shown that only a pair of vertically separated pressure probes can determine a single dominant spatial frequency of a fixed surface being touched, and also an arbitrary spatial spectrum of a moving surface profile without knowing its speed. Theory of vertical sampling in the elastic body is presented. Several experimental results utilizing a pair of pressure probes embedded in a silicone rubber sensor tip are shown.>